YC W26 Launch Video Analysis: What Drives LinkedIn Engagement
The difference between 3,521 reactions and 53 came down to three decisions. We analyzed 173 launch videos across all 179 W26 companies to find them.
By Guillermo Castro
Founder, Unveil
173
Videos Analyzed
111K+
Total Engagement
503
Median Engagement
3,521
#1: Pocket
Context
The Natural Experiment
All 179 launches posted from YC's own LinkedIn account
Same account. Same algorithm. Same initial push. Founder networks add a boost (r=0.31), but YC's channel does the heavy lifting. What really varies: the video itself.
Hook Analysis
We Scored Every Hook Twice
Once based on the visual opening — the first frames someone sees before the audio kicks in. Once based purely on the spoken words — what the transcript says in the first few sentences. Two independent scores. Both correlated with engagement. The findings lined up.
Visual hook rating
Manually scored 1–5 for every video: how well do the first frames stop a scroll? Covers visuals, energy, framing — everything before the audio. n=173.
1–2
326
median eng
81% flop
2.5–3.5
488
median eng
51% flop
4–4.5
658
median eng
33% flop
4.5–5
823
median eng
14% flop
r = 0.38
Spearman correlation with engagement. 2.5x median gap top vs bottom. Every tier higher = better performance.
Spoken hook (transcript only)
Independently scored the opening lines of all 166 video transcripts — words only, no visuals. Same 1–5 rubric: does the first sentence earn attention before selling? n=166.
1–2
355
median eng
67% flop
3
507
median eng
49% flop
4
543
median eng
40% flop
5
702
median eng
17% flop
r = 0.29
Spearman correlation with engagement. 2x median gap top vs bottom. Weaker than visual — but still significant (p < 0.001).
Both matter
Visual hook is the stronger signal (r=0.38 vs 0.29) — the first frame someone sees before the audio even starts matters more than the first sentence they hear. But both correlate independently. The best-performing launches get both right. The worst get both wrong.
What the Spoken Copy Actually Says
We pulled the transcript opening of all 166 videos and looked for patterns. What separates a top-scoring spoken hook from a bottom one isn't vocabulary or polish — it's structure. Two failure patterns showed up in weak hooks that almost never appeared in strong ones.
78%
"Hi, I'm [name]..." — flop rate
18 videos opened with a greeting then immediately the founder's name. 78% flopped — median 398 engagement. Shofo (1,269 eng) opened "Hey! Hey, 6-7! Against all odds..." — 8 seconds of scene before a name was ever said. Robby (841 eng) opened "Hi Jack, you're on your way to Mr. Connelly's house." — straight into a scenario. Put something worth watching before you say who you are.
260
median eng — product-description opener (vs. 503 batch)
When your first sentence IS what your product does, most flopped. "Introducing Fenrock, AI agents for back office at banks" (264 eng). "RIDAVAL helps regulatory teams submit FDA documents 12x faster" (260 eng). The one that worked: "Introducing Kofia, AI automations that write themselves" (Cofia, 1,848 eng — "Kofia" is the phonetic rendering in the transcript). The second clause was the hook — the unexpected detail. The rest answered "what does your product do?" before the viewer asked.
Side by Side: The Exact Transcripts
Here's what the top 5 and bottom 5 hooks sound like, word for word. Read them back to back — the difference is obvious.
Strong hooks (4.5–5/5)
Pocket — 3,521 eng
"This is Pocket. I never go anywhere without it."
GrazeMate — 2,464 eng
"I grew up on my family's cattle station in Australia."
Ressl AI — 1,530 eng
"We're building AI for the industries that Silicon Valley left behind."
Prana — 1,378 eng
"When you felt off after that long bike ride, when your heart skipped a beat..."
Chamber — 1,622 eng
"I need more Compute! Your teams are scrambling for GPUs like it's Black Friday."
Weak hooks (1–2/5)
Ditto Biosciences — 233 eng
"Hi, I'm Bader. I'm Dennis. And I'm Emily. And we're Ditto Bio."
Talking Computers — 101 eng
"The biggest bottleneck to AI co-workers joining the workforce is communication."
Fenrock AI — 264 eng
"Introducing Fenrock, AI agents for back office at banks."
ZeroSettle — 133 eng
"Zero-set-all is a mobile SDK that switches users from App Store billing to direct billing."
Crosslayer Labs — 53 eng
"I'm going to show cross-layer labs detecting a hijack attack on a financial services platform."
The pattern across weak hooks
Weak hooks give away the answer. Strong hooks create a reason to keep watching.
The two most common failure patterns: leading with your name (78% flop) and opening with a product description (57% flop). What the top performers share isn't a formula — it's that the first few seconds create curiosity, show a scene, or surface a problem. The viewer leans in before they know what you're selling.
Production Quality
The Quality Cliff
Editing
Excellent Editing (4.5-5)
874
median engagement
VS
Poor Editing (1-2)
283
median engagement
3.1x
Editing is the single biggest lever in the entire dataset. The gap between polished and rough is over 3x. Nothing else comes close.
Hook Strength
Strong Hook (4.5-5)
823
median engagement
VS
Weak Hook (1-2)
326
median engagement
2.5x
Your first 3 seconds make or break the video. A strong hook won't guarantee you go viral, but a weak one pretty much guarantees you won't.
Overall Quality Tiers
Excellent Overall (4.5-5)
597
median engagement
VS
Poor Overall (1-2.5)
236
median engagement
Video Length
Every Second Has to Earn Its Place
Length doesn't cause flops — filler does
The sweet spot is 45s–1:15 (626 median), but it's not a hard rule. Flop rate climbs as duration grows — 27% under 45s → 39% at 45s–1:15 → 58% at 1:15–1:45 → 68% over 1:45 — yet 39% of videos over 75 seconds still beat the batch median. GrazeMate ran 126 seconds and hit 2,464 engagement — 2nd highest in the batch. Longer videos don't fail because they're long. They fail because every second wasn't earning its place. Short videos with filler flop just as hard. Medians shown. 173 videos with duration > 0.
median engagement →
Under 45sn=15 · 27% flop
604
45s – 1:15n=64 · 39% flop
626
1:15 – 1:45n=66 · 58% flop
482
Over 1:45n=28 · 68% flop
454
Production Style
Style vs Engagement
Four production styles — each video tagged by what it actually is. Hybrid is the most common format at 68 videos and the strongest performer. Pure talking heads, with no motion graphics or b-roll, hit the highest flop rate in the dataset.
Best: Hybrid
582
median engagement
VS
Worst: Talking Head
346
median engagement
Hybrid
Talking head + motion graphics, often with screen demo — the most versatile format
582median eng · n=68 · 41% flop
Motion Graphics
Primarily animated — kinetic type, 2D/3D animation, illustrated explainers. Often combined with b-roll.
478median eng · n=49 · 55% flop
Screen Demo
Product walkthrough or screen recording as the primary visual — no founder on camera
392median eng · n=36 · 67% flop
Talking Head
Founder direct-to-camera only — no motion graphics, no b-roll, no animation
346median eng · n=37 · 76% flop
1.7x
Hybrid is the format of choice — 68 of 173 videos, the most common style and the strongest performer. The real differentiator within any style is cinematic quality: videos with high production value hit a 692 median engagement vs 485 for the rest. Pure talking heads — founder on camera with no motion graphics or b-roll — hit a 76% flop rate, the highest of any style.
The Cost of Getting It Wrong
What Bad Production Actually Costs You
We watched all 173 videos. The same mistakes kept showing up. These are the five that cost the most engagement.
The Length Tax: Going Over 2 Minutes
454Over 1:45 median
vs
62645s–1:15 median
Going long hurts. Videos over 1:45 pulled a 454 median vs 626 for videos in the 45-second to 1:15 range. But the problem isn't really length — it's filler. If every second earns its place, a 90-second video is fine. If it doesn't, people bounce.
The 3-Second Window: Weak Hooks
326Hook 1-2/5 median
vs
823Hook 4.5-5/5 median
Strong hooks hit 823 median. Weak ones hit 326 — a 2.5x gap just from the opening seconds. LinkedIn decides fast: if people scroll past the start, the algorithm stops feeding the video to more people. You don't get a second chance at that first impression.
The Editing Gap: 3.1x at Stake
283Edit 1-2/5 median
vs
874Edit 4.5-5/5 median
Editing mattered more than anything else we measured. Well-edited videos got 874 median engagement. Poorly edited ones got 283. That's not about fancy transitions or effects — it's about pacing. Cutting the slow parts. Knowing when to move on.
The Human Factor: No Faces on Screen
392Screen demo median
vs
545Other styles median
Pure screen recordings don't land. They pulled a 392 median vs 545 for everything else. But pure talking heads were even worse (346). What worked was mixing it up — some product footage, a founder on camera, a bit of motion graphics. Variety is what holds attention.
The Algorithm Blind Spot: Reactions Over Comments
17.1% Hybrid comment rate
vs
14.3% Motion graphics comment rate
LinkedIn cares way more about comments than reactions — roughly 5 to 10 times more. Hybrid videos pulled a 17.1% comment rate vs 14.3% for pure motion graphics. Videos that give people something to talk about get more reach than videos that just look polished.
Timing
When You Post Matters
No hour-of-day data — but day-of-week tells a clear story. Monday launches outperformed every other weekday. Wednesday was the worst. Weekend data is included below but too sparse to draw conclusions from.
median engagement →
Mondayn=27 · 41% flop
601
Tuesdayn=37 · 49% flop
521
Wednesdayn=38 · 61% flop
433
Thursdayn=36 · 50% flop
527
Fridayn=29 · 45% flop
529
Weekend — small sample, treat with caution
Saturdayn=5 · 60% flop
456
Sundayn=1 · 0% flop
897
Sunday: 1 data point only — median and flop rate are not meaningful here
Monday is the clear winner — Wednesday is the one to avoid
Monday posts hit a 601 median with the lowest flop rate of any weekday (41%). Wednesday is the outlier in the wrong direction — 61% flop rate and the lowest median at 433. Tue, Thu, and Fri cluster tightly in the 520s. Avoid Wednesday. If you can post Monday, do it.
Founder Network
Does Founder Reach Matter?
175
Companies Scraped
382
Founder Profiles
r=0.31
Spearman Correlation
A bigger following gives you a real 2.2x lift
Founders with 10K+ followers averaged 882 engagement vs 405 for those under 2K — a genuine 2.2x boost. That said, correlation is r=0.31, so followers still leave most of the variance unexplained. Video quality and hook drive the rest — and you can't buy your way past a weak video with reach alone.
Engagement by Follower Tier
< 2,000n=11
405
2,000 - 5,000n=71
509
5,000 - 10,000n=59
640
10,000+n=34
882
10K+ Followers
882
mean engagement
2.2x
<2K Followers
405
mean engagement
Takeaway
Growing your LinkedIn audience before launch day pays off — that 2.2x lift is real. But it's not a prerequisite. The Token Company had just 2,925 followers and hit 2,157 engagement with a top-tier video (4.1/5 composite) — a small account with a well-executed video can still beat the big-account average.
Team size? Doesn't matter at all (r≈ 0)
Solo founder or team of five — makes zero difference to how your launch video performs.
By Vertical
Industry Breakdown
#1
Hardware / Robotics
19companies
766median eng.
Top: Pocket, GrazeMate
#2
Climate / Energy
8companies
700median eng.
Top: DAIVIN!, Terranox AI
#3
AI/ML Infrastructure
15companies
610median eng.
Top: The Token Company, Chamber
#4
Developer Tools
38companies
526median eng.
Top: Cofia, RunAnywhere
#5
B2B SaaS
56companies
488median eng.
Top: Lance, Ressl AI
#6
Healthcare / Biotech
19companies
401median eng.
Top: Prana, Patientdesk.ai
Outliers
The 3% Exception
Low Quality, High Engagement (6 of 179 companies)
Only 3.4% of companies pulled this off. Each one had something unusual going for them — a viral product concept, perfect timing, or an existing audience. These are lottery tickets, not a repeatable strategy.
Cofia
1,848
Quality: 1/5
Chamber
1,622
Quality: 2.5/5
DAIVIN!
1,398
Quality: 1/5
Orthogonal
904
Quality: 2.5/5
Clam (formerly Baseframe)
897
Quality: 2.5/5
Agentic Fabriq
741
Quality: 2.5/5
Interactive
Every Company, Plotted
Directory
All 179 Companies
Every YC W26 launch video ranked by engagement. Includes post copy, video transcript, and frame screenshots for the top 30.
HookPocket (YC W26) is a small device that takes notes for all your conversations and meetings.
Every important conversation happens in person — investor meetings, doctor visits, client calls, team standups. And every note-taking tool on the planet only works over Zoom.
How It WorksFor in-person conversations, the best technology available was a pen. So the team built Pocket. Three mics. One button. You press it, have your conversation, and a summary is waiting for you in the app — full transcript, action items, and to-dos, pulled out automatically. You don't type. You don't take notes. You just talk.
SolutionPocket also does something no app can: snap it to the back of your phone and its contact mic captures both sides of a phone call. No speakerphone required.
TractionIn the last 5 months, the team has delivered over 30,000 units with a $27M annualized run rate, growing 50% month over month.
Congrats on the launch, Akshay Narisetti and Gabriel Dymowski!
🚀 heypocket.com/now
HookThis is Pocket. I never go anywhere without it. At the tap of a button, it records and creates organized notes based on your conversations.
ProblemI use it at work, at the doctor's, and when I'm out and about. I could use an app, but this, it's effortless. I snap it to the back of my phone and take it with me everywhere I go.
How It WorksI open the app and Pocket gives me full notes for everything I record. It knows who said what and action items are pulled out automatically. It even has a mind map of all my to-dos from every meeting in one place.
TractionIt also works offline, like fully offline. Record anywhere you want and it'll sync when you get back online. Thousands of people use Pocket.
CloseDoctors, lawyers, students, founders, basically anyone managing a packed schedule. This is Pocket.
HookGrazeMate (YC W26) builds autonomous drones that herd cattle.
ProblemLarge ranches can spend hundreds of thousands each year moving cattle between pastures using motorbikes, helicopters, and horses. It's expensive, time-consuming, and increasingly hard to staff.
How It WorksGrazeMate's drones herd cattle automatically. They find your herd, guide them calmly using AI that understands animal behavior, and provide real-time animal and pasture data. GrazeMate lets ranchers manage their operations on their terms - moving and monitoring cattle more often, with real-time insights to support each decision.
TeamSam Rogers grew up on a cattle station in Australia and dropped out of robotics to build robot cowboys. The GrazeMate team combines PhDs and backgrounds in agriculture.
CTAIf you're a rancher looking to save time and do more with less, schedule a demo at grazemate.com or send a message to sam@grazemate.com.
Congrats on the launch Sam Rogers!
HookI grew up on my family's cattle station in Australia. Most days we'd spend hours moving cattle between pastures so they have fresh grass to graze. This can be done on horseback, on motorbike and often by helicopters.
ProblemBut the reality is, it takes massive amounts of time. Ranchers can spend up to 5 hours every day moving cattle. It's expensive.
Operations can spend anywhere from tens of thousands to over a million dollars a year on mustering alone. And finally, it's dangerous. I've seen first hand how this can go seriously wrong.
SolutionThat is why we built Grazemate. Grazemate is an autonomous drone system that musters cattle for you. It learns from your herd and gives ranchers back their time.
Here's how simple it can be. Via your phone, you open the app, select where your cattle are and pick where they need to go. Then, you press start.
The drone takes off, it finds your cattle and helps move them to the new pasture. It reads animal body language in real time and adjusts its position to keep them moving calmly. Grazemate's AI learns how your specific herd responds and gets better with every muster.
Once it's done, you get a notification. There's no manual piloting and it's completely autonomous. Beyond mustering, Grazemate gives you complete visibility over your operation.
How It WorksThe same drone can monitor cattle health, estimate weight and check pasture quality along with inspecting water troughs. I've watched my dad spend hours every day moving cattle. Soon he'll do it with a few clicks over his morning coffee.
TractionAnd he's not alone. We're already working with ranchers managing over half a million cattle. And now we're bringing this across the world.
SolutionGrazemate is the new way to help ranchers manage cattle.
Excited to finally announce that The Token Company (YC W26) is a part of Y Combinator working on ML models to pre-process LLM inputs.
SolutionOur goal was to fix context bloat by compressing LLM prompts to make them shorter, denser, and cheaper to run.
We then realized the compressed inputs actually made the LLMs perform better.
TractionWorking with Pax Historia (60k DAU, 15th largest token consumer on OpenRouter, 193B tokens/month), we saw users prefer outputs generated from compressed inputs.
Pax Historia also saw a lift of +5% in user purchases after compression. They not only saved money but also increased user purchases by compressing inputs before sending them to the LLM.
We are seeing this effect now with several of our customers handling large natural language inputs
["Dance of the Sugar Plum Fairy"] ["Dance of the Sugar Plum Fairy"]
HookLance (YC W26) builds AI agents for hotel operations that manage guest communication, sales workflows, and on-property execution end-to-end. They're already working with leading hotel groups in the US operating 50+ hotels across major brands.
How It WorksHotels still run on phone calls for everything: towels, late checkout, maintenance, parking, and directions. Every request triggers a manual workflow, someone responds, gathers details, routes the task, tracks it down, and follows up. That model breaks when teams run lean. With 65% of hotels reporting staffing shortages, the front desk becomes the bottleneck. During peak periods or after hours, hotels miss up to 40% of inbound requests because the operation does not scale.
Lance fixes this by handling guest communication 24/7, qualifying and responding to sales and booking inquiries, and coordinating real on-property execution. The system collects the right details, routes work automatically, escalates when needed, and supports
HookY Combinator just backed us to solve an industry problem that has been untouched in the past 35 years. Some of the biggest hotel brands in the world are currently using REI Agent to run guest operations.
SolutionWatch how Lance takes a guest request from end to end, start to finish. Lance meets you where you are. It connects to your existing and legacy systems, platforms, and workflows.
In minutes, Lance is ready to answer a guest request and take actions on their behalf. Hey Lance, you live? Hey, what can I help you with today?
ProblemI'm thinking of hosting a pool party after demo day. Can I extend my stay a couple more days? That shouldn't be a problem.
DemoLet me check availability and extend your stay. Since it's after demo day, want me to reserve a restaurant near the pool and make sure extra towels are ready? Yeah, that's great.
Can we do a reservation for 8? Absolutely. I'm seeing availability at 8.
TractionI'll book that and arrange extra pool towels as well. Does that work? Perfect.
Thanks. All set. I'll text you the confirmations.
SolutionHave fun at the pool and good luck at demo day. Whether guests call or text, Lance handles both at the same time. It speaks their language in English, Espanol, Jongwen, Hindi, Francais.
ProblemAnd it follows your brand and your policies exactly. Every request is tracked from start to finish. Nothing gets lost and teams finally see what's slowing them down.
How It WorksHotel technology isn't moving forward, but Lance connects hospitality's outdated systems to give guests an experience that's truly seamless. Go to Lance.live because guests have places to be.
HookCofia (YC W26) creates AI automations that write themselves.
How It WorksCofia learns how you work and builds tailor-made automations you can deploy without prompts, code, or workflow builders.
Everyone talks about AI saving time, but the reality? Most AI tools require you to: recognize the opportunities for automation, describe your exact workflow in a prompt, and learn new interfaces and configure new tools. So repetitive work stays manual, even when AI could handle it.
Traditional AI waits for you to describe what you need. Cofia proactively builds personalized automations, trained on your exact workflows, by securely processing the work you actually do.
No prompts. No remembering to turn it on. No describing what you do.
How it works:
Solution- Cofia learns about the specific repetitive tasks you do
- When it recognizes a pattern, it builds a custom agent and then offers it to you
Problem- You review, launch it, and never do that task manually again
Sample tasks it'll detec
HookIntroducing Kofia, AI automations that write themselves. It securely learns how you work and proactively builds personalized automations that take work off your plate and do tasks just like you would. Take a sales rep, building prospecting lists, copying from the internet, pasting into a spreadsheet, enriching with search and Salesforce data.
ProblemAgain and again. Kofia comes to the rescue when you're doing it manually. It knows how you've done this in the past and is now offering you a custom-built automation for this use case.
SolutionReview it, launch it. Your automation now handles it all. 400 leads found with the exact process you usually follow.
Now take a recruiter, posting the same job to five boards. Copy, paste, format, repeat. Kofia says, hey, I know you want to post this in all boards.
How It WorksI can take over. All you need is one click. Review, launch.
Traction30 minutes saved per posting. Or a marketer, stitching together Monday morning reports from Google Ads, Facebook, HubSpot. Kofia knows you do this every morning and offers to help.
Now you can enjoy your coffee instead. The magic? You never wrote a prompt.
How It WorksYou never described your workflow. Kofia never asked. It learned and built agents tailored to your work.
CloseReady to see what it can take off your plate? Download Kofia today.
HookRunAnywhere (YC W26) helps apps run AI directly on phones and other edge devices. You add our SDK in a few lines, run things like chat and voice on-device, and fall back to the cloud only when you need to—so it stays fast, works with bad internet, and keeps sensitive data local.
ProblemThe problem today is that "on-device AI" is still a pile of sharp edges: every platform is different, models are hard to ship and update safely, and teams end up stitching together downloads, memory management, and multiple runtimes themselves.
How It WorksRunAnywhere is designed to take that burden off the app team: it handles model delivery and updates, keeps the app stable while models load/run, and gives you a dashboard to roll out changes and see what's happening in production.
TractionTheir team has already done this at scale—Sanchit has shipped mobile SDKs and apps used by millions, and Shubham has built large-scale infrastructure and reliability systems for massive fleets.
Congrats on the launch Sa
HookThe biggest threat to a data center is if the intelligence can be packed locally on a chip that's running on the device. If I can only run it on the mainframe, arguably the application is not that good.
ContextRunning on-device AI models faster than ever. The phone will really be an edge node for AI inference. My car lights are on, what should I do?
SolutionIf I can design it such that when the device becomes powerful, I don't have to redesign my application. Then I have built an application that is durable to the innovation that is undoubtedly going to happen.
You'll get everything through AI. Whatever you can think of, or really whatever the AI can anticipate you might want, it'll show you. How well can you actually run it?
CloseAlmost anywhere. And anywhere is the key word over there.
HookChamber puts your AI infrastructure orchestration, governance, and resource optimization on autopilot so that you can run on average 50% more workloads using your existing AI infrastructure.
ProblemThe platform acts like an autonomous infrastructure team, monitoring GPU clusters, predicting resource needs, detecting bad nodes, and reallocating GPUs in real time so AI teams can move faster without manual work.
Today, 30–60% of enterprise GPU capacity sits idle due to siloed allocations, driving up costs and slowing teams, adding up to more than $240B in waste each year.
TractionChamber is a team of former Amazonians that helped build and scale large-scale infrastructure optimization, delivering hundreds of millions in cost savings. Now they're bringing those learnings to help other enterprises.
Congrats on the launch Charles Ding, Andreas Bloomquist, Shaocheng Wang, and Jason Ong!
https://www.usechamber.io/
HookI need more Compute! Your teams are scrambling for GPUs like it's Black Friday. They wait months for a training slot and don't make progress on their models.
ContextEvery day, an Excel spreadsheet decides who gets to train. For you, Compute has turned into a back-alley economy. Looking for GPUs?
ProblemAnd you'd do anything to get your jobs running. But here's the twist. You probably have enough GPUs.
TractionIntroducing Chamber. Run up to 50% more jobs on GPUs you already have. Chamber puts your AI infrastructure governance and GPU optimization on autopilot.
We detect idle capacity in real time and shift it to the jobs that matter most. Bad nodes are handled autonomously. Jobs run uninterrupted.
ProblemMy model is finally training! No manual intervention. No more spreadsheet micromanagement.
HookCrow (YC W26) lets users control your app through chat. Connect Crow's AI agent to your product, and users can type what they want instead of clicking through menus.
ProblemSoftware UIs are stuck in the past. Users expect to chat with products like they do with ChatGPT, but building an AI assistant that actually works takes 6+ months. Meanwhile, support tickets pile up and users churn to new 'AI-first' alternatives.
TractionThat's where Crow comes in. Connect it to your product's features and data, and it executes real actions based on what users ask for. It navigates your UI, calls your endpoints, and handles complex workflows. You set the guardrails, track every interaction, and deploy in under a week.
Crow's customers go live in under a week - getting an AI Assistant on their app that is faster and more reliable than what they tried building in-house.
How It WorksAryan and Jai built Crow after transforming workflow-heavy products into chat-first experiences at top AI startups—they kno
HookRessl AI (YC W26) deploys AI Employees at Field Operations businesses to automate their office.
How It WorksEvery home services business runs on "glue" employees - people who spend their days copy-pasting between software, generating estimates, chasing parts vendors, answering calls, and manually reconciling invoices in QuickBooks. They're doing work on top of software that should be doing it automatically. The result: bloated offices, slow response times, and leads falling through the cracks - all eating into margins that were already thin.
TractionRessl is replacing that manual, grunt work with OpenClaw-style proactive AI agents that live inside the tools these businesses already use. Their agents handle estimating, procurement coordination, multi-channel lead response, insurance communications, and invoicing - autonomously, in the background, without anyone clicking through a janky UI.
Leaner office means higher margins. Every lead answered instantly means more jobs booked. Tha
HookWe're building AI for the industries that Silicon Valley left behind. The trades industry. While every other industry got an AI revolution.
ProblemAnd home services got service data and a player. SMA businesses are facing decade-old problems. Overwhelmed by notifications from all directions.
Estimates taking days. How far is this job? Is that part still in stock?
And the margin pressure? Don't even get me started. That's where we come in.
Meet Rassel, an AI employee that runs your front office for you. Rassel instantly gives estimates using passcodes. Books jobs and answers queries across any channel.
Dispatches the right technicians for the job. Follows up with your leads. Picks up calls 24x7 and gives you a full view of your business in one place.
How does it work? Rassel has access to past data from your software and all your messaging platforms. Which means it knows your business like you do.
All you have to do is shoot a text to Rassel like you would to your employees. And Rassel would act immediately. Rassel, what did you do last night?
CTAI got us 21 jobs. 3 on Yelp, 6 on email, 12 on SMS, and I sent estimates to all of them. If you're a services company looking to increase your operating margins.
Or a PE firm looking to transform your portfolios. Let's talk.
HookDAIVIN! (YC W26) enables humans to breathe underwater without oxygen tanks.
Did you know a single bottle of water contains enough oxygen to sustain a human for a day?
DAIVIN uses electrolysis to extract the oxygen from water and convert it into breathable gas for divers, enabling tankless underwater breathing limited only by battery duration.
Leo Kankkunen is an electrical engineer with seven years of experience in the electrical industry and the youngest person in Finland to achieve the highest national electrical certification. He's also a certified diver and previously served in the military.
ProblemUnderwater diving technology hasn't changed in almost 100 years, and is rife with problems.
SolutionSo DAIVIN! removes tanks entirely, enabling missions to start anywhere with a shoebox-sized system.
Congrats Leo Kankkunen on the launch!
www.daivin.tech
I don't know if you are the guy that took this video from the water, but let's try it. Here you can see the device. Let's do it. Here you can see the device. Let's do it. Let's do it again.
Thumbnail
3 sec
Mid
Final
#11
Galactic Resource Utilization Space, Inc. (GRU Space)
HookGRU Space (YC W26) is building off-world habitats starting with a hotel on the Moon. Learn more and apply to stay at gru.space/reserve.
GRU develops in-situ resource utilization technology, transforming local lunar material into durable building structures. The first habitat is a lunar hotel for space tourists, targeting opening in 2032. From there, GRU aims to lay the foundation for America's Moon base, future Mars cities, and a truly interplanetary civilization.
This launch comes at a pivotal inflection point in human history. For the first time, becoming interplanetary is no longer science fiction. With the U.S. announcing plans for a permanent Moon base by 2030, the next decade will define humanity's expansion beyond Earth, and GRU is building the infrastructure to make it possible.
TractionSkyler Chan, graduated early from Berkeley EECS to found GRU. Previously he built vehicle software at Tesla, built a NASA funded 3D-printer launched into space, authored at the wor
HookTensol creates autonomous AI employees, powered by OpenClaw.
How It WorksMost AI agents still require you to prompt them, lose context between sessions, and can't work across your tools. OpenClaw changed that, but deploying it securely for companies is a nightmare.
Tensol deploys AI employees in a secure, isolated environment with one-click integrations to your tools, audit logs and persistent memory of your organization. Set up takes 5 minutes.
SolutionCompanies are using Tensol employees across their business - support agents that resolve tickets and fix bugs, SDRs that follow up with leads and keep your CRM updated, operations agents that chase invoices and create reports, and more. They work 24/7 and can remember everything.
Congrats on the launch, Oliviero Pinotti and Pratik Satija!
HookOpenClaw Everybody's using OpenClaw, but setting it up for companies is a nightmare. We fixed that. Hi, I'm Oliviero, and we're building Tencel, AI employees for your company built on OpenClaw.
SolutionWe recently discovered OpenClaw and understood the massive potential it can have for companies. So we built Tencel to help every company enable it in a couple of clicks. Let me show you our AI employees in action.
ProblemImagine you're in a meeting. Suddenly, you get 10 customer messages that the checkout button is broken. Dev, your AI engineer, who's constantly monitoring your sentry logs, fix it up, spins up cloud code, fixes it, and pushes it.
TractionNow comes the magic. He also does the handoff to Sophie, your AI customer support employee, who immediately sends messages to all of the 10 customers affected by this issue and notifies them. She has intricate context about every customer and can pull data from the CRM to personalize the messages.
All of this with 10 customers, two AI employees, all deployed within less than five minutes. With Tencel, you can deploy your AI employees in secure environments in three simple clicks. You can give your agent a persona, an AI model, and a heartbeat.
How It WorksThen you simply create your agent, and they work 24-7, even while you're asleep, while running in secure environments with end-to-end encryption. And finally, you can connect all your tools so your new AI employees have access to context across your entire organization.
CTAAt Tencel, we believe the next generation of companies will have their own AI employees. Try it out now at tencel.ai. He never checks out.
HookPrana (YC W26) is the 24/7 AI doctor in your pocket.
Traditional primary care is reactive. Between 15-minute annual checkups, subtle health declines go unnoticed. Because health data is siloed, doctors lack the context needed to catch metabolic and chronic diseases early.
SolutionPrana fixes this by unifying your medical records and wearable data for "always-on" monitoring. Their AI tracks your heart rate, sleep, and activity in real-time to detect risks before they become crises. When the system flags a concern, you're notified instantly and can escalate to a board-certified MD for immediate intervention. Through Prana, you can chat with a physician, get prescriptions sent to your pharmacy, and receive lab orders for deeper bloodwork — turning passive data into active medical care.
CTAPrana is officially live! Sign up today and join in building the future of proactive medicine: www.pranadoc.com
Congrats on the launch Meer Patel, Sanjit Menon, and Vishvam Rawal!
HookWhen you felt off after that long bike ride, when your heart skipped a beat getting ready for your late night shift, when you're home with your crushing headache waiting for your appointment in three weeks, these are the moments that matter, moments healthcare system misses. We cannot go on like this.
SolutionWith Prana, the 24-7 AI primary care doctor, your health has continuous oversight. Prana monitors your SmartWatch health data for shifts in heart rate, sleep, and activity, helping you catch issues early.
On your commute, chat with Prana about your health and symptoms, like that back pain that won't go away. When you'd like to escalate care, you can chat with a board-certified doctor for diagnoses, prescriptions, and labs, so your life doesn't have to pause because of preventable health issues.
CTAStay on top of issues early, stay productive, and stay healthy. Prana.com For more information visit www.pranahealth.com Prana.com
HookCardboard Inc. is an agentic video editor that lives in your browser.
How It WorksUpload raw footage, tell it what you want, and get a first cut in minutes. You get the power of tools like Premiere Pro or DaVinci Resolve without the learning curve.
ProblemMost teams sit on tons of raw assets (talking heads, screen recordings, B-roll, product demos) but ship far fewer videos than they should. Not because of lack of ideas, but because editing is slow, manual, and kills momentum.
SolutionCardboard fixes that. You can ask for a cut, search footage by meaning, and refine it yourself on a timeline. Think of it as vibe editing.
If you use video for launches, ads, recaps, or weekly content and care about speed, try Cardboard.
Congrats on the launch Saksham Aggarwal and Ishan Sharma!
HookYou've got hours or days of incredible footage. And you have the perfect story floating around in your head. But here's the problem.
ContextYou're spending 90% of your time just finding the right clips. Dragging them around. What if you had a friend who knew all your footage inside out?
SolutionIntroducing Cardboard. An AI video editor that lives in your browser. You can literally search your footage by description in plain English.
Ask it to make a 60 second vlog. Done. The music doesn't hit.
CloseGenerate an original soundtrack that actually fits your vibe.
HookHuman Archive (YC W26) builds datasets to model sensorimotor intelligence for robotics and world models. HA-Multi is the largest dataset of manual labor tasks aligning egocentric RGB, stereo depth (IR dot projection), tactile gloves, body IMUs, and wrist cameras.
SolutionHuman intelligence is predominantly embodied, and the internet does not capture it. Every day we perceive the world, apply force, and adapt to noisy real-world environments. Yet no dataset captures this multimodal structure of human interaction with the world at scale. As a result, progress in embodied spatial intelligence has been bottlenecked.
TractionTo collect data at scale and ensure quality, they build and deploy custom hardware across homes and businesses, then post-process the resulting data through internal QA, anonymization, annotation pipelines, and policy benchmarking.
CTAThey've already shipped data to frontier research teams and will open-source samples next week. If you're interested, reach out: raj@h
HookToday, we're introducing Human Archive. We're building the foundational datasets required to model human-sensi-motor intelligence at scale. The past decade of AI was about modeling languages, image, and video.
SolutionBut the next frontier is capturing the structure of human interaction in the physical world. We're engineers from Stanford and Berkeley, and over the past two months, we've set up the infrastructure to collect data at scale, including our custom hardware, internal models for benchmarking data quality, and our amazing 25-person ops team.
We've been live for less than one week, and now we're actively scaling our datasets. Human Archive Multi is a fully aligned, multimodal dataset that includes vision, stereo depth, tactile, motion, and wrist cameras.
Human Archive Ego is a mono-RGB wrist camera dataset. We index heavily on diversity, collecting data across residential homes, restaurants, hotels, construction sites, logistics, and industrial environments worldwide.
TractionFor our customers, we provide structured outputs and visualizations, including 3D mano-hand reconstructions, 2D tactile force maps, human pose reconstructions, and depth maps for every timestamp. We also provide post-processing and annotations, like hand tracking, action labels, SLAM, object segmentations, and full 3D pose reconstructions.
SolutionBecause we are able to collect and synchronize this 3D data as the ground truth, many of our annotation models are able to perform at state-of-the-art levels. For embodied agents to truly generalize, the scalable path forward is not to constrain data collection to the current action space of robots, but to instead expand models and hardware to absorb naturalistic human data as it exists.
And egocentric video alone is not enough. Each sensing modality represents a unique dimension of interaction, and only through their alignment can we begin to approximate the underlying structure of human embodied intelligence.
TractionTo do this at scale, we design our hardware so it seamlessly fits into workflows. Upon arrival of our upgraded hardware, we intend to scale data collection by orders of magnitude. Next week, we're open sourcing the largest multimodal dataset of its kind, and we're just getting started.
CloseOur work will expand far beyond robotics and world modeling into scientific research on the nature of intelligence itself. Let's archive the world for embodied intelligence.
HookMaywood automates deal execution for investment banks, from decks to diligence.
How It WorksDeal teams still juggle CIMs, diligence, models, and buyer feedback across fragmented workflows. Nothing talks to each other across data rooms, spreadsheets, slide decks, and endless email threads. As a result, dealmakers can't focus on what they're best at: closing deals.
SolutionMaywood reinvents the deal process with a single, living workspace for every deal.
How It WorksThe Maywood platform automatically generates finished, on-brand CIMs, presentations, models, and diligence responses that prioritizes editability and auditability, with best in-class security ensuring your information stays protected. When a number changes, everything updates instantly. Every material stays synced and ready to send. Faster processes and tighter execution mean more deals and better outcomes.
SolutionDrake Goodman, Kent Goodman, and Esteban Vizcaino built Maywood after seeing firsthand how much deal momentum gets lost between
HookThere was $5 trillion of global M&A last year, but the deal process hasn't fundamentally changed in over 20 years. Your materials are in one place, diligence in another, and buyer's feedback scattered across email.
ContextNothing talks to each other. Teams don't lose deals on rigor. They lose them in the chaos between systems.
SolutionUntil now. Introducing Maywood, the first platform to run an entire deal. Maywood leverages your firm's specific patterns, historical data, and analysis of what makes your pitches win.
We generate the materials that evolve with the deal instead of being rebuilt from scratch. These aren't drafts you clean up. They're finished deliverables your MD can send tomorrow.
Retrieving information is easy. It's producing work at the level your clients expect that's hard. We solve that.
How It WorksWe built Maywood to be auditable, editable, and dynamic to fit inside your team's workflows. Diligence runs in the same system. Questions answered, tracked, and tied back to the materials.
When facts change, all deal materials update instantly and automatically. A living data layer that evolves with you from pitch to close. Unlike chatbots, Maywood produces the real work.
The decks, the models, the diligence responses, the way that your firm actually delivers it. And then some. Your deal materials shouldn't take longer than your flight to the management meeting.
Book a 30 minute call today to modernize your workflows.
HookShofo (YC W26) is building the Common Crawl for videos.
ProblemAI labs need massive video datasets, but high-quality, segmented video data is hard to access.
How It WorksShofo is assembling the largest index of short-form videos. They build complete pipelines that collect, segment, sanitize, and label videos from across social media to curate custom datasets for AI labs.
TractionFor example, if a lab needs 50k cooking videos featuring hand-object interactions, Shofo queries their index, runs it through their labeling pipeline, and delivers a clean, annotated dataset.
The founders previously built Correkt, an AI search engine for multimodal content that reached 43k users. They've spent years developing proprietary infrastructure to collect and index videos at scale.
Want data? Contact them at founders@shofo.ai
Congrats on the launch Bryan Hong, Andre Braga, Braiden Dishman, and Alexzendor Misra!
HookHey! Hey, 6-7! Against all odds.
ContextSay that we've been beefing, dawg. Remember that famous story? Idiot.
TeamHi, my name is Brian Hong. I dropped out of Berkeley to build Shofo, the world's largest index of short-form videos. We're collecting, cleaning, and labeling every post on social media to curate them into data sets that power the next generation of AI models.
Traction75% of the world's videos are locked away on social media platforms. Shofo unlocks this. We collect billions of hours of video data from these platforms, cleaned and segmented by hundreds of attributes such as creator demographics, audio information, transcripts, and engagement patterns, and then index it all into a queryable database with embeddings for semantic search.
For example, AI labs can search for makeup videos narrated in Spanish with 500,000 likes, whose top comments mention natural or glowing, and get a curated data set of exactly that. And finally, as a teaser, we're launching the largest short-form video data set ever released onto HuggingFace.
CloseCheck out Shofo.ai and train your models on us.
HookCascade (YC W26) makes AI agents reliable and safe by learning directly from how they behave in real workflows.
How It WorksEvery company runs differently so artificial intelligence must adapt to each organization's workflows, but most agents today are effectively static after deployment. Teams iterate prompts, add rubrics, and manually inspect logs without a reliable way to determine whether behavior is aligned with intent. Failures and threats are difficult to attribute, and performance quickly plateaus.
Cascade addresses this by building custom evaluation infrastructure that learns from production runs. Judgments can then be leveraged to generate training signal from a company's own operational data to continuously make agents safer and more reliable on their workflows.
Cascade's goal is to unlock latent intelligence across an organization by enabling effective agentic deployment. They're building adaptive scaffolding that can operationalize across different domains. Casca
Hook🎵 Sun is shining in the sky 🎵 🎵 There ain't a cloud in sight 🎵 🎵 It's not raining, everybody's in the play 🎵 🎵 And don't you know, it's a beautiful day 🎵 🎵 Hey 🎵 🎵 Running down the avenue 🎵 🎵 See how the sun shines brightly 🎵 🎵 In the city, on the streets where once was pity 🎵 🎵 Mr. Blue Sky is living in the day 🎵 🎵 Hey 🎵 🎵 Mr.
ContextBlue Sky, please tell us why 🎵 🎵 You had to hide away for so long 🎵 Hey. How are those benchmarks going? All done.
Identified all the failures in our past runs, annotated them, auto-populated the evaluator. Now to find two models running in production. And, uh, what's going on over there?
Those guys? Yeah. They ran their agents for that cascade.
HookAutumn AI (YC W26) is building the first real-time signal intelligence platform for GTM teams.
ProblemToday, teams have to piece together buying signals scattered across posts, tweets, commits, and blogs. But these signals decay fast, and manually monitoring dozens of sources per person across thousands of prospects is nearly impossible.
TractionAutumn tracks prospects across the internet, at scale. GTM teams define their ICP and the signals that matter, and Autumn delivers a condensed, real-time feed filtered by intent. Autumn is already powering workflows across industries, from fintechs prospecting newly incorporated tech companies to deep-tech companies gearing for pre-conference outreach.
Congrats on the launch Vishnu Sampathkumar and Shiv Kampani!
HookHi, I'm Vishnu. And I'm Shiv. We spent the past four years together at Columbia Building.
ProblemRight now, your prospects are telling the world exactly what they want to buy. On social media, in product updates, in public conversations. But you're missing these signals.
Today, we're excited to introduce Autumn, the world's most comprehensive signal intelligence platform. Let's say you're trying to sell into an enterprise. A traditional lead gen tool would give you a massive list and say good luck.
TractionOur scouts would tell you that a VP of engineering at that company just crashed out on X about her data pipeline. Or that two engineers just left her team. Now she's panic searching for a new solution.
And that's your signal. With Autumn, it's all in one live feed. Tell us what you care about and we'll track real-time signals from LinkedIn, GitHub, X, conferences, incorporations, filings, etc.
Our customers are currently using us across industries. From DevOps, to manufacturing, to banking and finance. We're live at automai.com
HookRoboDock (YC W26) builds robots that run autonomous depots for autonomous fleets.
ProblemAutonomous fleets can drive themselves, but they can't operate themselves.
By 2035, there will be ~76M autonomous vehicles globally, with the AV market projected to grow from ~$200B today to ~$3T. AVs are scaling fast, but the depots that charge, inspect, and maintain them are still manual. That means labor-heavy operations, costly downtime, and expensive infrastructure buildouts.
How It WorksRoboDock deploys a drop-in robotic system that retrofits onto existing chargers to physically plug vehicles in and run automated post-trip inspections using vision and thermal sensing, no rebuilds required.
By automating charging and inspections, RoboDock turns depots from manual cost centers into self-running, scalable infrastructure. This is how autonomy starts becoming truly end-to-end.
TractionZinny and Celine have deployed autonomy across sky, highway, and home — from drones to Class 8 trucks to consume
HookAutonomous vehicles aren't coming, they're already here. They run over 20 hours a day, and soon, millions of people will trust them with their lives. But the systems that keep them running, still completely manual.
ProblemAfter each trip, vehicles return to depots, where crews charge them and inspect them under tight timelines. That's thousands of manual actions every day, millions in operational costs per year. As fleets grow, this becomes impossible.
This is RoboDoc. RoboDoc turns depots into self-running systems. Our system knows which vehicle arrived, what it needs, and when to charge.
How It WorksIntegrated vision and thermal sensors verify every check, before the vehicle is cleared to drive. Every RoboDoc learns, every depot gets smarter. This is physical AI.
Now this isn't just a concept, we're co-developing with key AV operators. No replacements, no retraining, just a drop-in upgrade. This is what autonomy looks like, at scale.
HookBusiness teams waste 100s of hours every week manually extracting, copy-pasting data and tab-switching across disconnected systems. APIs are missing, systems are fragmented, and traditional RPA breaks the moment a UI changes.
RamAIn (YC W26) kills the busywork by teaching AI to move data between legacy systems, desktop applications and web-portals by simulating mouse and keyboard clicks just like a human.
How It WorksRamAIn learns the UI structure of your key interfaces and uses your team's computers 10x faster and more reliably. Takes days to setup rather than weeks. Its self-healing to UI changes, supports human-in-the-loop oversight, maintains a full audit trail, and combines API integrations with UI automation into a unified agent.
If your team can see it, RamAIn can automate it.
CTAJoin logistics, insurance, procurement, healthcare and finance teams automating workflows with RamAIn today → https://ramain.ai/automate
Congrats on the launch Shourya Vir Jain and Vansh Ra
HookYour company's data lives everywhere, across desktop apps, web portals, and legacy systems. And when you think of automation, it stops where APIs end. Meet Raman, an AI agent that uses your computer just like a human, moving data between web portals, legacy systems, and desktop apps, ten times faster.
ContextInvoice processing is a breeze with Raman. It starts by learning your interface, like a human. It then navigates to your partner portal, finds an approved invoice.
It scans and reads the data inside, extracting everything into clean, structured data. Then, it navigates to your ERP and inputs the relevant information. It visually navigates across the interface, ensuring it doesn't miss a single detail.
SolutionIt handles drop-downs, dynamic data types, and scattered input fields. Once done, it confirms the final amount and relevant details, and the invoice is sent. Done.
How It WorksRaman connects internal apps with external apps, no matter how they're built. One unified agent for all automation. Smarter.
Faster. Works anywhere. Automate workflows no one else can with Raman.
Right now, a proposed regulation is moving through a committee somewhere that will cost your company millions. You probably don't know about it. Your policy team (if you have one) might not either.
SolutionBut Fed10 does. Their AI agents ingest every proposed regulation, flag threats to your business, draft the fix, and show you who to call and what to say to get it done. In seconds, not weeks.
TeamThe founders passed bills before they could even vote. They led immigrant advocacy campaigns in Nevada and got bills signed into law that now affect 344,000 Nevadans annually. They dropped out of Harvard, Williams, and Berkeley to build this.
Congrats on the launch, Armand Iorgulescu, Winston Wei, and Zayan Islam!
🇺🇸
HookHey, I'm Armand. Hi, I'm Winston. And I'm Zion.
ContextAnd we're ex-lobbyists who passed legislation before we could even vote. And now we're building FedTen, AI agents that help companies learn about regulatory threats before they cost them entire markets.
Last August, Illinois passed a single law that completely wiped out the AI therapy market. The companies building in this space, they only found out about it after it was too late to act. And things like this happen all the time.
Big companies spend $30 billion a year on lobbyists and consultants to still get blindsided by regulatory changes. And that's because no team can keep track of hundreds of thousands of policies across the globe.
How It WorksBut FedTen can. So here's how it works. FedTen learns from you.
DemoWe build company context by looking at public filings, industry data, and internal documents to create a custom legislative agenda for you. We then look at every active jurisdiction on a federal, state, and municipal level to find the bills that affect you.
But we don't just give you a list of bills. We tell you exactly how each one affects you and whether it's a threat or an opportunity. And when you need to write amendments to change them, we can help you with that as well.
We're a full-stack government affairs software who can even get you the right lobbyists. We've been in policy since we were teenagers. We passed legislation before we could even vote.
TractionAnd that's how we know the system is broken. Companies still pay consultants hundreds of thousands of dollars a year for intelligence that's still too slow. FedTen delivers in seconds what consultants deliver in weeks.
CloseWe believe that every business deserves a seat at the table. Stop being reactive to legislative changes. Instead, be proactive with FedTen.
HookSkillsync (YC W26) helps companies find elite (but overlooked!) engineers on GitHub based on what they have actually built. It analyzes public GitHub contributions and turns them into structured skill profiles that recruiters & hiring managers can search.
SolutionSkillsync is uniquely valuable for hard engineering domains like low-level systems, robotics, or infra, where real signal lives in code. Many of the top candidates in these fields don't maintain LinkedIn profiles.
The team behind Skillsync experienced this problem firsthand while running a large open-source project with hundreds of contributors and more than 20,000 stars. When they tried to hire their most elite contributors full-time, they discovered how tedious and manual it was to identify and evaluate talent directly from GitHub.
TractionSkillsync is the answer. Instead of keyword matching on resumes, it lets you search directly for capabilities, shortlist engineers already building what you need, and engage them wit
Everyone claims they code these days, but how can you quickly find the greats?
HookStilta (YC W26) is building the Cursor for patent practitioners.
How It WorksInnovation is key to solving the world's challenges, but the tools to protect innovation aren't keeping up. Patent practitioners are stuck doing high-stakes work with processes built around manual review and disconnected point solutions.
Until now.
TeamStilta replaces rigid tools with agentic AI for deep patent and competitive research. Every output is interactive, argument-led and traceable to cited passages. The founders are four engineers from McKinsey & QuantumBlack that spent years implementing secure AI solutions for the most demanding Fortune 500s.
TractionStilta is already live with leading firms in Europe and US, ensuring nothing is missed when it matters the most.
Congrats Oskar Block, Oscar Adamsson, Petrus Werner and Tobias Estreen on the launch!
For more UN videos visit www.un.org Like and subscribe for more!
HookAurorin CAD (YC W26) is building new professional CAD software from the ground up to be extremely fast and AI native. A part that takes an experienced SolidWorks user 20 minutes to make only takes seconds in Aurorin.
SolutionCAD is how engineers design satellites, robots, cars, and so much more, but CAD design is painfully slow and tedious. All major tools use the same couple of slow CAD kernels built in the 80s. It's common for these tools to take 4 hours to open complex assemblies.
Aurorin has rebuilt CAD software from the ground up with their own kernel. This unlocks a much better AI interface AND software speed than previously possible.
Before starting Aurorin, Michael Baron worked at SpaceX on Raptor combustion simulation, Dragon guidance, navigation & control, and Starshield flight software. He also worked on GPU driver performance at Apple. With this combination of skills, he is building the future of smart and fast CAD software.
Congrats on the launch, Michael!
HookEngineers build satellites and rockets and robots and so much more that the world desperately needs. Engineers use CAD software to bring ideas to reality. The problem is CAD is painfully slow and tedious.
SolutionAuroran is fixing that. We built new CAD software from the ground up making a new CAD kernel designed for rapid execution speed and to be AI native. Here's what that speed looks like in practice.
This part took about 20 minutes to make in SOLIDWORKS. In Auroran, just a single natural language prompt in seconds. Auroran is full mechanical CAD software with sketches, constraints, assemblies, mates, and all the other features you'd expect.
With Auroran, you can work in natural language or standard CAD UI. Its AI agent can generate 3D parts with constrained sketches, complete with GD&T tolerances through model-based definitions, and then it can generate your engineering drawings for you.
CTATeams are already switching from their existing CAD software to Auroran and others are using it for rapid prototyping alongside their existing tools. If you would like your engineering team to move way faster, reach out now.
HookServo7 (YC W26) builds robots to automate industry work.
How It WorksToday, industry automation requires a full redesign for a rigid, single-purpose robot to operate. That complexity makes implementation slow, expensive, and risky, and it prevents many businesses from automating at all. That's a missed opportunity.
Servo7's wheeled humanoids and robot arms are trained with AI to operate directly in existing workflows. They learn from a small number of demonstrations, then continue improving on the job to work faster over time. This shortens deployment and increases flexibility.
TractionThe team has experience in deploying autonomous systems in demanding environments, and are already working with warehouses and CPG brands.
Congrats on the launch Pieter Becking and Jasper van Leuven!
HookYou're wasting millions of dollars on custom industry robots that take months to implement. Hi, I'm Pieter. And I'm Jasper.
ProblemAnd we are here to fix that. Today, implementing robotics means redesigning your whole facility so the robot can function. This is inefficient and unnecessary.
With Server 7, you don't have to do any of that. Our robots adapt to your environment. Here's what it looks like in practice.
How It WorksLet's say you have a conveyor belt with products on them, and an employee who picks the products from the conveyor belts and puts them into boxes. Now, you want to automate this process. Instead of adjusting the floor plan, redesigning the conveyor belts and the outflow, Server 7 robots just work in your existing process.
TractionWe're already working with CVG brands and e-commerce warehouses to automate their fulfillment. We've worked on autonomous defense systems and surgical robots. Not just in the lab, but actually deploying them in the field.
CTALet our robots handle the repetitive work. If you're working in logistics, assembly, or manufacturing, we would love to work with you. Reach out to us at founders at server7.com.
Hooksitefire (YC W26) is building GEO on Autopilot to improve your brand's AI visibility.
SolutionUnlike other GEO tools, sitefire tells you how to improve your visibility in ChatGPT & co. It recommends actions, and takes them for you. They are already working with Europe's largest brands such as BMW, Xtrackers, and DWS.
Tractionsitefire identifies which content drives AI visibility and tells you which third-party sites to engage with, and what content to create. sitefire analyzes the top-cited pages for your topics and creates a brand-aware, AI-optimized version of them - delivered directly to your CMS.
CTATry it out at sitefire.ai
Congrats on the launch, Jochen Madler and Vincent Jeltsch!
HookHi, we're Vincent and Jochen and we just raised half a million dollars from YC to build Sightfire, the future of how brands are discovered in AI search. Unlike other GEO tools, Sightfire actually improves your visibility.
ProblemIt recommends actions and it takes them for you. As of today, more than a billion customers are using AI agents like ChatTPT. These agents don't serve links, they provide answers.
And if your brand is not in that answer, you don't exist. This is where Sightfire comes in. Sightfire identifies which content drives your AI visibility.
How It WorksFirst, it tells you exactly which third-party sites to engage with. Second, Sightfire uses this content to tell you which blog posts to write. And with one click, it writes them for you.
TractionSo you can improve your visibility and start winning links, built by data scientists from Stanford and TUM and already trusted by Europe's largest brands. Don't just write, be the answer. Sightfire.ai
HookTerranox AI (YC W26) is the first AI-powered uranium discovery company.
ProblemNuclear is experiencing a global resurgence. 30+ countries have pledged to more than triple capacity by 2050, Microsoft, Google, Amazon, and Meta have collectively committed billions, and the U.S government signed executive orders to quadruple capacity with $80B+ committed. But this future has one critical dependency: uranium fuel. Global production needs to 4x by 2050, yet we already have a supply deficit today. The world's largest mines will approach end-of-life within the same decade, and with traditional exploration hit rates below 1%, we're not discovering new deposits fast enough to close the gap.
SolutionTerranox finds uranium deposits in North America using AI models trained on decades of exploration outcomes. Their system unifies fragmented geoscience data, identifies high-potential targets, and optimizes every exploration decision to maximize information gain per dollar spent. Every drill hole
HookGoogle announcing a deal to purchase nuclear energy. Constellation Energy said that it will reopen the Three Mile Island nuclear power plant to Microsoft. One of the most significant corporate purchasers of nuclear energy in American history.
ProblemEvery reactor needs one thing to run, uranium, and the world is already burning it faster than it can dig it up, and demand is about to explode, but the fuel's out there, hundreds of meters underground, undiscovered until now. Our models are trained on decades of exploration data, focusing in on the most prospective regions, narrowing to the most promising land, determining exactly what to do and where to drill, and then we prove it in the ground.
I'm Liav. And I'm Jade. The world needs uranium to power the future.
HookOrthogonal (YC W26) is building a simple way for AI agents to discover and pay for APIs.
How It WorksAI agents are about to become the biggest consumers of APIs, but today's API ecosystem was built for humans. Agents have to manage accounts, API keys, auth flows, and billing across dozens of services, which makes using paid APIs slow and unreliable. As a result, many agents can't access the tools they need, and API providers leave money on the table.
Orthogonal fixes this with a single MCP or SDK that gives agents instant access to paid APIs. No API keys, no accounts, and no subscriptions to manage. Just pay per request. API providers list once and become discoverable to every agent on the platform.
Christian Pickett previously worked on payments at Coinbase and billing at Vercel. Bera Sogut worked on reCAPTCHA and Maps APIs at Google. Between the two of them, they lived in payments and API infrastructure, and this problem sits right at the intersection.
If you're building
HookHi, I'm Christian. I'm Bera. And we're building Orthogonal.
SolutionAPIs weren't built for agents. We fixed that. AI agents are about to become the biggest consumers of APIs on the internet.
How It WorksMCPs that require payments and identity means managing accounts, API key, billing, across dozens of services. The API economy was built for humans, not agents. That's why we built Orthogonal.
Orthogonal is where agents go to find and pay for APIs. I have a sales meeting with Kier Marosh. I want to do deep research and find as much information as I can about him.
The agent is searching for and finding many different APIs like X profiles, LinkedIn profiles and different search providers. Excellent. Now I have everything I need for my call with Kier Marosh, his life, his career, his company.
If you're building agents and want easy access to APIs or if you have APIs and want to be discoverable by every agent on our platform, contact us at founders at orthogonal.com.
HookHaladir (YC W26) is building operational superintelligence.
Operationally-complex industries like logistics, critical software, and manufacturing make thousands of decisions a day under tight, overlapping constraints. Current models fail here. They hallucinate, give confidently wrong answers, and can’t reason through systems with millions of interdependent moving parts.
How It WorksHaladir connects LLMs to formal constraint solvers, giving models a fundamental understanding of constraints rather than just context. The team builds at both the model training and application layer, and is currently working with a leading foundation model company.
The founders published in IEEE and Elsevier Q1 journals for OR and ML in high school, and left Carnegie Mellon, Princeton, and UVA to build Haladir.
Congrats on the launch, Jibran Hutchins, Joseph Tso, Quan Huynh, and Preston Schmittou!
https://lnkd.in/g5Uyk5eN
HookThe biggest thing holding AI back from being deployed in mission critical systems is verifiability. For high stakes domains like critical software, logistics and operations, a confidently wrong answer is worse than no answer at all.
ProblemAnd over the past few years, deploying LLUZ and enterprises have failed to deliver real returns, not because models lack intelligence, but because operating inside constrained systems requires more than just semantic context. And to solve this very issue, today we are launching holidayer, the operational super intelligence company.
How do you use an AI product research lab that looks on one thing, making AI reliable and effective for real-world operational decisions? We are improving AI at the model training and application layer, making LLUZ accountable for formal reasoning.
How It WorksToday, LLUZ is on cable, and I can open the decisions under real-world constraints, limiting their ability to make decisions as scale. The opportunity is not to replace these models. It's a connect them to improving constrained solvers and formal methods to reach me.
TractionThe intersection between these two technologies will allow AI to finally be deployed to make critical decisions. And we're just getting started.
HookData security is a major issue among AI frameworks like OpenClaw
Clam (YC W26) brings enterprise-grade security to such AI systems through their “Semantic Firewall”, a security checkpoint that sits around the AI’s environment at the network level.
ProblemTheir system runs a series of scans on anything your AI sends out into the world, and everything that comes in, to catch inconsistencies before they turn into problems.
Before Clam, Anshul and Vaibhav worked at Series B startups leading AI observability and evals for enterprises, and building secure remote coding agents.
How It WorksWhether you’re an enterprise or an individual exploring AI, your data should stay protected. Deploy and set up OpenClaw built on security with Clam.
Congrats Anshul Paul and Vaibhav Agrawal on the launch!
https://lnkd.in/g7tUzUPS
HookHey everyone, this is Uncholco founder of Clamp. To show you how our security works, let me first show you on the platform. Once you create an account, you'll begin chatting with Clamp as it guides you through different use cases and setup.
ContextWe've added a series of integrations and use case templates to get you started quickly. Once you create a task, you can view and manage it to see where your agent is up to and add more as expanded list of to do so.
Let me also add a task to summarize all resumes I've gotten through email overnight. Now let's say it's 8 a.m. and the air wants to read data from my inbox.
Clamp security sits on the network layer and audits the traffic between your AI instance and the outside world. Think of it like a firewall, but for your AI. Your AI doesn't need to see all your data to function, just what it needs for what you've told us to do.
CloseThat's where we come in. Thank you. Check us out at triclamp.com.
OctaPulse (YC W26) uses robotics and computer vision to automate fish inspection and processing for aquaculture farms.
HookAquaculture is the fastest growing protein sector on earth. See food demand keeps rising and production is scaling fast. Fish convert feed to protein six times more efficiently than cattle.
ContextA fraction of the land, a fraction of the emissions. When you're feeding 10 billion people, that math and matters. Production needs to grow 30% by 2030 to meet global demand.
TractionAchieving that requires better tools and labour alone can't scale fast enough. Aquaculture is a $350 billion industry and the challenge isn't just growing your fish. It's getting the most out of every gallon of water, every pound of feed, every square foot of the farm.
We take fish inspection from five minutes to under 30 seconds with over 95% accuracy, automated phenotyping, grading and quality control. Today, we're live with a six-figure contract with the largest trout producer in the US.
CloseAnd this year, we're beginning global expansion. From hatchery to harvest, we're building the operating system for aquaculture. We're Octavols and we're helping feed the world.
Arzule (YC W26) uses AI to turn partnerships into a more predictable and scalable source of revenue for B2B SaaS companies.
HookPartnerships drive nearly 40% of software as a service revenue, but behind the scenes, there's still manage and spreadsheets and driven by intuition and gut feeling. Kind of like how we chose these outfits today.
ProblemWe're back by WICOMNator to build a more automated data-driven revenue. Starting with predicting and scaling the B2B partnerships that actually drive revenue. Today, companies don't know which partnerships will truly matter, or where they have leverage.
SolutionThe part of their life cycle lives in spreadsheets or basic PRMs, and everything else is manual gas work. Who'd reach out to, when to follow up, and what's actually driving results? We built our Zool to bring intelligence to ecosystem, expansion, and growth stage B2B SaaS companies.
TractionWe work with companies, already dragging over $400 billion in annually recurring revenue through partnerships and help them scale it faster. For Zool users specialized, always on research agents, to continue a seem modern and a licensed data sets and the public web, are Zool to analyze his market trends and ecosystem signals to identify which partners will actually drive revenue before it becomes obvious.
Instead of guessing who might better, you see exactly where leverage exists. Once a partner is identified, our Zool helps structure the relationship, automate outreach, and measure ROI attribution across touchpoints, events, and co-marketing.
It continues to recommend where to focus, so that you invest in the partnerships that compound. With our Zool, you can send and receive partnership requests, build trust through past collaboration history, and alert high-intank customers inside your partner's ecosystem's that perks you offer.
Our Zool is available as software, or as a fully managed partnership growth engine, or re-identified partners, initiate conversations, and structure high-leverage relationships on your behalf. While others are still building partner lists, our customers are already closing revenue through them.
Button Computer (YC W26) is the wearable AI that can talk.
HookOkay, Bun. Farthery. Left or right side.
ContextOn the left, close to the heart and done. I'm looking for both the combines by the finance and the level of the fantasy. What about tax go for dragons by MJ Shalakwoo?
Done! How do I know when this bad boy is ready? In turn, I'll temp 130.
So, by the sounds of it, it may be well done. Done! How do you say piece of junk in Germany?
DemoYou do this bloody schrub thing. I'm going to fist you in need for one attack. Depends on the size of the taco.
Robby (YC W26) turns every technician visit into additional revenue.
HookHi Jack, you're on your way to Mr. Connelly's house. He asked for a filter replacement and to do a tune-up.
ContextHis boiler is 12 years old, so we're taking a look whilst you're there. He's a premier customer of ours and prefers shoes on. Does that work Jack?
Got it. Thanks, Robbie. Hey Robbie.
Hey Jack, how did it go? Just finished off at an air flow issue. This is up to run in pretty clean now.
I also replaced the air filter while I was in there. Anything the office should follow up on? Yeah, I checked the boiler.
TractionCustomers open to the employees. I'll let the office know. Thanks, Jack.
Kita (YC W26) turns messy financial documents into fraud-checked, predictive risk signals that power lending in emerging and underserved markets, starting in Southeast Asia.
HookIn emerging markets, a borrowers financial truth lives inside messy documents. Banks statements, business financials, pace slips, and more. Lenders spend weeks piecing together noisy data and unexpected formats.
SolutionBut now, the era of manual review is over. Introducing Kida. The first document intelligence layer built for lending and emerging markets.
Kida uses vision-language models to read real-world documents, detect fraud, and extract high confidence underwriting signals in seconds. It understands local document patterns where generic models break, then links signals to real repayment outcomes.
Getting smarter with every loan. The result is faster approvals. Lower risk.
Cut costs. An accredited engine that compounds with scale. Mue Kita, turn documents into signals.
Chasi (YC W26) deploys AI agents that help equipment dealers sell more, respond faster, and maximize fleet utilization - 24/7.
HookThat's what every equipment dealer rich take a do. Hi, I'm Akash and I'm sorry man. We built tractors, race cars, robots and AI at phones like Tesla and Boy.
ProblemNow we are bringing AI to the equipment industry. Right now your team is drowning in calls, takes emails and contracts. Request slip all the time your equipment goes idle and the industry loses like $300 billion every year.
SolutionSo we asked, what if we gave every dealer an AI admin an AI that gives your team superpowers introducing chassis? Let me show you how it works. When a request comes in, phone email, web form, or text, chassis collects the information, reouts it to the right person, and helps them respond even after hours.
TractionWith full context about the customer, what they bought and what they probably need instantly. And after the response, the CRM and ERP are already updated. chassis understands where every customer is from first contact to contract.
How It WorksIt recommends what to do next, not just your rectifier for the web, or reaches out automatically. And if not just following up on request, chassis practically reaches out for contract renewals, service appointments, paymail reminders and preventative maintenance.
CTABut here is what it gets interesting. Because chassis has context from your CRM, ERP, conversations, IoT data, public data and website visits, its surface is insights to help your team maximize revenue, and even helps your team prioritize the right accounts and improve their sales approach.
CloseThat's chassis, the AI platform that gives your team superpowers so they can grow, retain market share and maximize utilization, tried for 90 days, risk free at chassis.low slash pilot.
HookForum (YC W26) is building the first regulated exchange to trade on cultural attention. Learn more and try the beta today at forum.market.
SolutionAttention drives billions in value across media, brands, and markets. Until now, there's been no way to hedge or manage cultural risk — even as companies bet massive budgets on sustaining online relevance. Forum's exchange changes that, enabling businesses and investors to trade relevance for the first time.
How It WorksForum builds an Attention Index by aggregating engagement signals — views, likes, comments, and shares — across search, streaming, and social. You can go long or short on this index, profiting from the topics shaping culture today.
SolutionOwen Botkin (CEO) traded long/short equities at Balyasny; Joseph Thomas (CTO) built software at GEICO and NASA. Together, they're creating a new asset class.
Congrats on the launch, Owen and Joseph!
https://lnkd.in/gvYV3kpE
HookTwenty-26 kicked off with a bang. What happened? This is crazy.
ProblemThe Epstein files dropped. More than three million pages. The internet went into overdrive, but the attention spikes started hours before the headlines.
La Boubu went from blind box toy to billion dollar phenomenon. Marty Supreme broke A24 records. Marty Supreme is an American film that comes out on Christmas Day.
TractionMachia took over every cafe and created thousands of new ones, all because of attention on social media. Bad Bunny just made Grammy history. You saw all of it coming.
What if you could have invested in it? Form is building an entirely new asset class. We track the online engagement of every and any topic you can imagine.
So now you can trade those viral trends you see coming before anyone else. It's not another prediction market. It's the next stock market.
HookPatientdesk.ai (YC W26) is a patient-aware AI booking system for dental clinics. It handles inbound calls, appointment booking, payments, and real-time insurance verification during the call, helping clinics capture more revenue without adding front-desk staff.
Dental clinics lose revenue at the front desk every day. Calls go unanswered after hours, insurance questions delay bookings, and generic AI tools fall back to callbacks when conversations become complex and patient-specific. Patients do not wait and high-value treatments are often lost before an appointment is ever booked.
TractionPatientdesk runs patient intake end-to-end by understanding who is calling and what they need, verifying insurance in real time, booking the right appointment, and handling payments directly inside the clinic's existing systems. The company was founded by Oncel, San, and Emre, who met over ten years ago in a high-school dorm and have worked closely together ever since, bringing years of hands-on experience working with clinics and front-desk workflows.
Congrats Oncel Ozgul, Fikri San Koktas, and Emre Kaplaner!
https://lnkd.in/ggj-_UN3
HookSarah. Hey Sarah. Go pick up the phone.
ContextThis is Sean. He tried to reach his favorite hair transplant clinic before the YC launch video. No one picked up.
I knew YC launch video was coming up, so I wanted to get my hair lined on. I call my favorite clinic. The line just kept bringing and ringing.
So I knew what to do. Just book a fly to Turkey, get it done, came back. And the phone was tear-ringy.
I just had this really bad tool take last night. I couldn't book an emergency appointment, so I just handled it myself. Handle it out.
You know, the player. Here kidding. I wish.
ProblemThis is Emre. He has had neck pain since last summer. He still hasn't been able to book an appointment.
So I've had this back pain for some time. And I decided to give my doctor a call. And when they picked up, it wasn't a person.
It just told me to press 6, 7, and then it thanked me for my patience. And told me they would call me back. This was last summer.
I'm still waiting. I'm still waiting. I'm still waiting.
I'm still waiting. I'm still waiting. I'm still waiting.
Health care has had the same bottleneck for decades. Why see back those to fix that? We are patient deaths.
TractionThey're already live across more than 50 clinics in the US, Australia, and the UK. And here is a problem. People call because they need help.
But clinics are busy. They don't want to drink. No one picks up.
When patients cannot get true, they do not wait. They move on. That's where patients are lost.
That's where revenue leaks every single day. That's not all. Clinics spend tens of thousands on ads every single month.
But follow up with their leads. Are the next day or after the weekends. The result is a list of phone numbers that no longer has.
We exist to fix this. We are an AI-booking system built specifically for healthcare practices to capture more people. But we go beyond answering phones or booking appointments.
We are on patient intake and to end. Without it in pressure to staff or changing how clinics operate. In December alone, we made one of our customers stranded at $50,000 in revenue with patients as booking zone.
How It WorksWe did not guess our way here. Before building patient deaths, we worked with clinics for years. We built and tested clinical workflows.
We trained from these teams. We ran our own call center. We saw where teams actually broke.
There are tools that can answer your phones. There are tools that can book appointments. There are tools that can verify patient insurance.
But nothing brings everything together and handles the patient intake completely. That experienced shaped patient deaths. We connect everything into one class called intake, insurance, payments.
TractionWe're already integrated with all major clinic management systems across the US, Australia and the UK. So when something needs to be booked, checked or updated, it happens automatically on the clinic systems.
But patients asked about coverage or payment. What would mine out of pocket cost be for this treatment? We check it during the call.
Great question. Let me check that out for you. Patients get answers right away.
CTAClinics do not have to call back later. After the visit, patient deaths keeps working. We follow up with the patients and when a payment is due, we call the patient and handle it without staff having the chance.
ProblemThe front desk stops being a bottleneck and starts working the way it should. Patient deaths. Patient deaths.
CloseBecause healthcare should not break at the front desk.
HookProtent (YC W26) is turning every camera into a real-time crime prevention tool, built for law enforcement.
ProblemToday, a single operator is responsible for watching thousands of live feeds simultaneously—body cams, drones, and street cameras. The result? Fewer than 1 in 200 incidents are caught in real time. Cameras are currently treated as retroactive tools, but they should be saving lives.
How It WorksFounded by Srihan Balaji and Abhisheik Sharma, Protent automates this by analyzing every feed at once. It instantly detects predictive threat signals and escalating situations, giving dispatchers and officers the immediate context they need to prioritize alerts and intervene before a situation spirals out of control.
TractionProtent is already enabling police departments and security teams to permanently shift from reactive response to proactive prevention.
Congrats on the launch, Srihan and Abhisheik!
https://lnkd.in/gVzDuy_n
Music Humanize weren't built for a thousand angles Music Music Detect a threat before it becomes a headline, deploy protest.
HookLucent (YC W26) is the AI that watches your session replays 24/7 to automatically detect bugs and UX issues - delivering alerts with full reproduction context to Slack and Linear.
ProblemMost users don't report bugs. They just churn. Teams collect thousands of session replays but nobody has time to watch them, so critical issues go unnoticed.
How It WorksLucent fixes this. Connect your existing session replay provider like PostHog, and Lucent monitors every session replay in real time, surfacing the bugs your users are hitting in realtime.
TractionThey're already used by top product teams including Reducto, Julius, Finta, and Happenstance. You can try Lucent for free: lucenthq.com
Founded by Alisa Rae, who previously sold an Australian EdTech startup and was employee #2 at MagicBrief (acquired by Canva).
Congrats on the launch, Alisa!
https://lnkd.in/gHCGhxj3
HookYour recording thousands of user sessions every day. But here's the thing no one talks about. Nobody's watching them.
SolutionBoxs are hiding in your products right now. Users hit them, get frustrated and leave, and you'd never even know. Introducing loosens.
How It WorksAn AI that ultimately watches your session replays and finds bugs and UX issues for you. Business connects to post-lock and start washing every session. When a user hit a bug, loosens the text to end sends it straight to your Slack and linear with a full bug record.
TractionOver 30 YC companies are already using loosens, to catch bugs they never find on their own. Your session replays are their vicious dollar source you're ignoring. Listen what's your system, so you can focus on building products people love.
DemoYou can set up loosens in one minute. That's it for the same time. Take care for you to watch this video.
CloseAnd you can try for free today at loosenshq.com.
HookRemy AI (YC W26) builds AI-powered robots for e-commerce fulfillment.
How It WorksIf you run a 3PL, you know the pain: SKUs are constantly changing, order profiles shift with every new brand you onboard, and peak season hiring gets harder every year. Yet over 80% of warehouses still prepare orders by hand: because existing automation is expensive, rigid, and takes years to see ROI.
Remy AI is building flexible robots to change this, starting with packing stations. Their robots handle variable SKUs, drop into existing pack-out workflows with minimal modifications, and pay back within weeks.
TeamOscar previously advised Fortune 500 logistics companies at BCG on warehouse operations and Ben studied machine learning as a PhD candidate at Oxford.
Congrats on the launch Oscar Brisset and Ben Kaye!
https://lnkd.in/guuiJvY9
We're MIA. And we're helping you build the warehouse of the future.
HookREV1 (YC W26) turns 3D CAD models into production-ready engineering drawings in minutes.
ProblemMechanical engineers spend over 25% of their time creating 2D drawings for the parts they design. It's tedious, manual, and absolutely required – if it's manufactured, it needs a drawing. It's also error-prone, and unlike software, you can't just push an update. Hardware mistakes mean costly redesigns and delays.
The worst part? Engineers aren't innovating during this process. They're just checking a box so manufacturing can begin.
SolutionREV1 changes how this gets done. They capture the engineer's design intent, then reason through the best tolerances and controls to ensure part functionality. Engineers stay in the loop to review and validate before sending drawings to vendors.
TeamThe founding team combines hardware and software expertise. Alex Rivero shipped products at scale as a mechanical engineer at Tesla and Apple, and lived this problem firsthand. Louis Liu built Tesla's global Supercharger planning platform and has published ML research. Together, they understand how hardware gets built and how to automate complex workflows with AI.
CTAIf you're looking to automate your drawings, reach out at: https://rev1.so
Congrats on the launch Alex Rivero and Louis Liu!
https://lnkd.in/efPVeP5h
HookMost people think this is how engineers build things. Reality is far from it. For example, the car that you drive every day, your favorite vacuum cleaner, and this guy, each one of his products is made out of thousands of individual parts.
ProblemAnd sure, engineering starts with 3D cut. That's the fun part. But actually, to bring things into production, every single part needs a 2D drawing.
And it can take more than 25% of the engineer's time. After done hundreds of these drawings at Tassel and Apple, and trust me, it's brutal. Every drawing needs dimensions, tolerances, service fish, manufacturing notes.
And they all have to follow this over 300 pages of rules. They are one symbol wrong, one tolerance off. That's a part that does not fit.
A machine that does not work. A part of college delayed again and again. Our R1 return 3D CAD is to production ready drawings in minutes instead of hours.
Not by talking to a chatbot. Communicate your design and tense directly in 3D. Then we generate the drawing.
Fuse dimensions, GDNT ready in minutes. Review each annotation with AI explanations, tweak, and approve. When it looks ready, export as PDF or back to your CAD PLN.
This doesn't just make engineering faster. It changes what engineers get to spend their time on. Less time drafting and more time solving the problems that are actually models.
HookOne Robot (YC W26) builds world model-based simulations that are realistic to see and realistic to interact with, so robotics teams can train and evaluate VLAs without being bottlenecked by robot time.
SolutionToday, improving a robot policy often means spending more hours in the real world: set up the scene, run trials, reset everything, record failures, tweak the model, and repeat. That loop is slow, expensive, and hard to scale. The founders of One Robot ran into this issue themselves while training robots for precision assembly work. At one point, evaluating a new policy started taking longer than training it. Existing simulators don't fully solve this because they usually miss one of two things: visual realism or interaction realism.
One Robot builds task-specific world model environments from robot data, so teams can run more training and evals, find failure modes faster, and iterate on policies with less dependence on real-world testing. They're starting with hard manipulation tasks like textiles and box folding, where realism really matters. The founders have been on the other side of this problem at Tesla, Google, NASA JPL, and robotics startups, and they're building the tool they wish they had.
Congrats on the launch, Hemanth Sarabu and Elton S.!
https://lnkd.in/db_eUYRv
HookWe don't have robots doing our chores today, because robots learn from experience and there aren't enough of them out there collecting data and running tests. If you wonder robot being proved today, you need more hours in the real world.
ProblemThis means setting up the scene, running trials, resetting and repeating. This is slow, expensive and impossible to scale. Robots and robots demand orders of magnitude more data and evaluations than what's possible today.
SolutionThat's why you don't see robots working with us, giving us back our time. Today's simulator are the bottleneck to that future. They don't look real or behavioral, so the robots can get experience they need to improve.
One robot fixes that. We build simulation environments that are realistic to see and realistic to interact with. So teams can iterate without being bottlenecked by robot time.
We do this by collecting data for the task we want to simulate. This could be textile manipulation or box folding and then we train a world model that can predict how the scene changes when the robot acts.
We have been on the other side of this. We've been training robots since 2020, a Tesla, Google, NASA, GPL. We kept hitting the same rule.
Real world testing takes longer than training models. That's why we're building one robot. We move fast.
TractionWe're already doing pilots with multiple robotics companies today. If you're training robot arms for material handling and logistics or manufacturing, let's talk.
HookPollen (YC W26) is building AI agents that make every customer feel like they're your first.
How It WorksCustomer success teams today are stretched thin - managing 40+ accounts each, toggling between dashboards, and manually stitching together signals from email, support tickets, and product data. As a result, churn risks go unnoticed and expansion opportunities are left on the table. And the only way to keep up has been to increase headcount.
Pollen fixes this by giving every customer account its own AI agent. Pollen connects to your email, support tickets, product usage, and CRM to detect real churn and upsell signals 24/7. Then, Pollen tells your team exactly which accounts need attention today and what to do next, letting you take action directly on the platform.
No more missed signals. No more busy work. No more scaling headcount just to keep relationships alive.
Congrats on the launch, Noah Yin, Jeffrey Yum, and Aldrin Ong!
https://lnkd.in/gNwjxT_b
HookEveryone remembers their first. That first customer stayed through broken features and bugs all because they felt like they mattered. But as company scale, every customer no longer feels that same level of attention.
ProblemImagine your biggest account hasn't logged in for weeks. The support tickets tripled last month. The customer success has no idea yet because they're buried in outdated dashboards trying to make sense of 40 other accounts.
SolutionSo your strong relationships that prevent a turn start breaking down. That's where we built Paul and. So every customer can feel like they're your first.
TractionPaul and gives every customer account its own AI agent. We connect to your email, support tickets, product usage data, and external news to detect real, turn and up sell signals 24.7. Then Paul and Axe automatically, fighting exactly which accounts need focus today and what to do next.
Like prepping the perfect email, QBR deck or renewal strategy. You can also travel with Paul and to get instant answers about things that would normally take hours. Like which accounts are most likely to discern this quarter or what wins should I highlight to this customer?
CTANo more busy work, no more miss signals, no more scaling head count, just to keep relationships alive. Book a demo at Paul and.cx to make every customer feel like they're your first.
HookEvery following we've ever uploaded is a protocol designed in 1974 until today. Introducing Dark by Blackquirt. Dark is a brand new protocol built from the ground up to accelerate viral transfer.
SolutionDark has outperform every public and proprietary transfer protocol ever tested against it. The speed alone isn't what we built this. Accomi built the CDN to push data to the edge.
We built the missing inverse, the world's first content ingestion network. Whitequirt is capable of going content in from anywhere on the internet, all powered by dark. With a door data lives on a robot, laptop, or cloud storage, whitequirt can reach it.
How It WorksSwap your endpoint URL and every upload is accelerated. Or connect any two file systems and let us handle the rest. Just a few clicks gives your data all these benefits globally.
CloseAre you a cloud AI or robotics company ready to ten extra data speed? Discover, whitequirt.
HookQ2Q (YC W26) helps deal teams find acquisition targets faster.
SolutionQ2Q is an AI native deal sourcing service for private equity, search funds, and independent sponsors. You tell Q2Q the types of companies you want to buy, and it'll get qualified meetings booked for you.
ProblemRight now, proprietary sourcing is extremely manual and requires an army of analysts to keep it running.
How It WorksQ2Q is your team of AI analysts that automates this process. Tell them your thesis, and they'll find targets and run personalized outreach to get intro meetings set up. They'll qualify the leads so you can just focus on the first call.
CTAToday, Q2Q is helping deal teams save time and source off-market opportunities. Want to try it out?: https://www.tryq2q.com/
Congrats on the launch Brianna L. and Claire Wu!
https://lnkd.in/gpMUBgda
HookWe just got back by YC to solve one of the largest problems in private equity, deal sourcing, search funds in private equity firms are ready using us to find potential acquisitions. And now we're excited to go live.
ProblemToday sourcing is slow. You open dozens of tabs, stock LinkedIn profiles, and write thousands of emails. Use three or four different tools, and it takes weeks just to get your first intro.
Wear your AI in terms that get your first calls both with companies you want to buy. Tell us about your thesis, and we'll put together a list of companies for you. Behind the scenes, we'll enrich each company with relevant data, contacts, and how likely we think they are to sell.
How It WorksWe'll automatically add companies that match your thesis to an outreach list. Now let's get your meetings booked. Our AI in turn can do email or phone outreach.
We'll go out and research each company on the web, socials and news, and write up a personal message for each firm. Or we can generate an outreach campaign through multiple platforms and with follow-ups.
As the human in the loop, you can review the messages before we launch. And you're ready to go. Every email reply or phone call is auto-track in our CRM.
We'll keep everything related to this prospect in one place, so you'll never lose any important information. You tell a cute accused of types of companies you want to acquire, and we'll get first meetings booked for you.
CloseIf you're in private equity or run a search fund, you know how tedious sourcing is. Want to meet profit faster? Go to trykitaq.com.
HookSentrial (YC W26) is the production monitoring platform for AI agent behavior. They help engineering teams detect, diagnose, and fix AI agent issues in production.
How It WorksAI agents are becoming core infrastructure - handling customer support, automating workflows, and making decisions on behalf of users. But when something goes wrong, teams are flying blind. Traditional monitoring catches errors and latency, not failures like hallucinations or user frustration. No alert fires when an agent goes off the rails, and agents that were meant to scale operations end up requiring constant oversight.
SolutionSentrial gives engineering teams real-time visibility into how their agents actually behave in production. They semantically detect failure patterns: loops, hallucinations, tool misuse, and user frustrations the moment they happen. When issues surface, Sentrial diagnoses the root cause by analyzing conversation patterns, model outputs, and tool interactions, then recommends specific fixes.
TractionWhether you're building internal copilots or customer-facing agents, Sentrial helps you run them reliably in production. Check them out at sentrial.com.
Congrats on the launch Neel Sharma and Anay Shukla!
https://lnkd.in/gGvXjE5Y
There are moments these AI models fundamentally have a technology by which they're predicting what's next and they are prone to errors. And I just think that it will take about decade to work through all those issues.
HookBalance (YC W26) is the AI-powered accounting and bookkeeping firm for forward-looking companies.
ProblemToday, companies still wait weeks for books to close, pay several thousands a year for basic bookkeeping, and are in the dark regarding the accuracy of their finances. The systems and accounting firms of today weren't designed for real-time business decisions.
SolutionBalance fixes this by combining AI agents with in-house finance experts. Their system pulls in context from your entire current finance stack like Quickbooks, Xero, Stripe, payroll, and banks - then reconciles continuously, flags issues, and responds to finance questions like a teammate who already knows your business.
How It WorksEvery close and return is reviewed by a human accountant, but the heavy lifting is fully automated. Clients get tax-ready books that are always up-to-date, without the cost or delay of a traditional bookkeeper.
TractionBefore Balance, the team built vertical AI application layer companies, scaled AI systems at global firms, and led finance ops for high-growth startups. Now they're building the accounting company they always wished existed.
CTABook a demo if your accounting is still stuck in the past. https://lnkd.in/eAQUdv_k
Congrats on the launch Gus Levinson, Emil Munk, and Mathias Løvring!
https://lnkd.in/es6PD6T8
HookWhen things are in balance, they just work. It feels effortless, not thinking about the invisible systems at work. Your counting should feel exactly the same.
ContextMinimal touch points, quiet confidence, everything handled in the background. That's balance. The accounting firm rebuilt for the AI era.
SolutionHere's what working with us looks like. Connect to your existing systems to balance one time. Then everything starts working automatically.
Receipts get pulled from your email automatically with AI. Expenses get categorized. Books get reconciled in real time.
Everything stays tax-ready. All happening in the background. No extra input needed.
TractionHave questions? You can ask our AI agent B anything. Revenue metrics, outstanding invoices, tax obligations.
B answers instantly. On Slack. What's up?
SolutionOr email. Balance learns how your business works and adapts. You are patterns.
Your preferences. When something needs human judgment, our accountant steps in. They do monthly check-ins, understand your business, and do more than your books.
They're part of the team. Because AI handles repetitive work, we offer a fixed monthly price at a fraction of traditional rates. No hourly surprises.
The same expertise. Just smarter execution. That's balance.
TractionYour accounting partner rebuilt for the AI era.
HookEmdash (YC W26) is an open-source desktop app that lets developers run multiple coding agents in parallel—without the chaos.
It is an Agentic Development Environment (ADE): one place to start tasks, dispatch agents from any provider, review what they changed, and ship.
Working with coding agents is still awkward. You kick off a task and wait. If you try to run multiple agents on the same repo, you end up with branch mess, merge conflicts, and a dozen terminals. And when a new model or tool drops, there's no clean way to compare it against your current setup on the same task—you're bouncing between your IDE, GitHub, and command line just to keep things moving.
How It WorksEmdash defines the workflow around coding agents. Each agent gets its own isolated Git worktree, so nothing touches the main directory. You can hand an issue from GitHub/Linear/Jira (or just a prompt) to one or more agents, run them in parallel, compare diffs in one place, make any quick manual edits, and create a PR.
SolutionEmdash natively embeds all popular agent providers so you can try the latest capabilities without retooling your setup. Arne Strickmann and Raban von Spiegel built Emdash for themselves, now they're turning it into the tool they wish existed for anyone building software with coding agents.
Congrats on the launch Arne and Raban!
https://lnkd.in/gHdWppnd
HookHi, we're Aban and Arna and today we're really excited to introduce you to M-Dash. In open source, a Gentec development environment. Let me show you who we use M-Dash to build M-Dash.
ContextI'm creating a new task and passing a linear issue to Codex. Codex immediately starts working on it. While the agent is running, I can check in on other tasks that other agents are working on in isolation.
ProblemHere you can see all the tasks that I'm currently working on. For example, in this task, the agent has complete running. I can see all the changes, edit them, revert them, or comment on them.
How It WorksOnce I'm happy, I can create a PR, a draft PR, or merge it directly. I can also test the changes by running a live stack ascript. Here in another task, you can see that we directly integrate the CI checks into our UI.
M-Dash comes with a file editor, a skills integration, MCP support, and further, we allow access to remote code basis via SSH. This is how you can shop software with M-Dash. Okay, three, two, one, go.
CTATry it on today and start shipping within this.
HookCopperlane (YC W26) builds AI-native software for end-to-end mortgage origination.
How It WorksRight now, loan intake is slow and manual. Borrowers upload the wrong documents, and loan officers spend days chasing missing info instead of closing deals.
With Copperlane, loan officers can process over 2x as many loans. Their AI agent Penny proactively closes loans by optimizing rate pricing, directing borrowers, and verifying docs. Loan officers receive complete files and can focus on bringing in new deals.
TractionCopperlane is already working with lenders processing hundreds of millions in volume annually. They'd love to help you too!: copperlane.ai
Congrats on the launch, Athan Zhang and Brianna L.!
https://lnkd.in/gJWasBCC
HookBuying a home is one of the most important decisions most Americans will ever make. And the process behind it, it's broken. Mortgage origination is a $2.2 trillion market, and this is a software lenders are still using to power it.
ProblemToday, getting a mortgage requires dozens of checkpoints between borrowers, loan officers, processors, and underwriters. It costs the lenders nearly $12,000 to originally a single loan, and loan officers spend more than half of their time chasing documents, asking questions, and filling out forms.
SolutionSo that's why we built copper lane. Mortgage origination is long overdue for disruption, and we're already working with lenders processing over $50 million a month in volume. Hi, I'm Ethan.
I was at Princeton and also part of the founding team for two previous startups. And I'm Brianna, the CEO. I went to work in working finance.
Ethan and I both come from mortgage families so that industry is deeply personal to us both. Copper lane is a-ionative mortgage origination software, and Penny is our AI agent that serves you. Penny sits at the center of the loan process.
She replaces the chaos as a single point of contact. Barers can talk to Penny 247 through our point of sale, over text, email, and over the phone. She answers their questions, guides them through the application, and follows up to the loan officer doesn't have to.
As Barers submit documents, Penny reviews them in real time. She flags mismatches, such as reported income that doesn't line up with a W2, and proactively reaches out to the Barer for an explanation before the loan officer has to.
Penny chases missing documents? Hi, Sarah. I noticed we're still missing your January bank statement.
TractionCan you upload it here? It allows us to automatically and push us loans from application to clear to close. We help loan officers handle twice as many loans and reduce loan processor workloads by 80%.
And Barer gets a port around the clock. 82% of Americans say home ownership is part of the American dream, but the process to get there hasn't changed in over a decade. Copper lane is the AI native platform that makes it faster, cheaper, and less painful for everyone involved.
CloseWe're a copper lane, and we're powering the path home.
HookProximitty (YC W26) automates complex loan servicing and collections so banks recover more, faster.
How It WorksLoan servicing and collections are still highly manual, costly, and full of compliance risk. Most external agencies recover less than 10% of outstanding debt, and most AI agents only handle narrow tasks, lacking the context to manage real workflows.
Proximitty gives banks and fintechs a no-code studio to build AI agents that automate end-to-end servicing and collections. This includes hardship requests across multiple systems, financial spreading, compliance QA, skip tracing, voice agents that take payments, and more than 100 other standard processes. It works with existing systems and workflows, so teams don't need a full overhaul.
The result is lower operating costs and more capital available for new lending.
CTABook a demo at www.proximitty.ai/
TractionWye Yew led Growth at Taptap Send ($50m -> $200m ARR) and previously advised banks and fintechs on credit risk at McKinsey. Zi led Bloomberg's security infrastructure that powers over 300k terminals and was Head of Engineering at ACI.dev (where he built the world's first unified MCP with 4k GitHub stars in 1 month).
Congrats on the launch, Wye Yew Ho and Zi Zhang!
https://lnkd.in/gqqZEa5i
Hookservicing in collections are heavily manual today in banks and vintex. And the cost of servicing in collecting a delinquent loan has risen drastically in the last five years. He's straining the flexibility to originate more than us.
SolutionSo banks are hiring more collections in servicing ages. Local origins of drowning, jumping between five to ten different systems, just resolve a single case, which is a huge constraint to lending. We've built proximity to fix this.
How It WorksWe unify all your lending data into one place. Then let you build AI agents that automate collections in servicing with zero engineering work required. Deploy voice agents that are personalized to each borrowers' case automate compliance monitoring and scale your teams' capacity without adding a headcount.
HookVela (YC W26) is the AI Scheduling Assistant built to handle the ambiguity that breaks scheduling.
SolutionMost tools fail when things get messy. Vela understands you: it knows to prioritize a client over an internal catch-up, and it understands "Let's do it early next week" without setting specific parameters. You just add Vela to the conversation, and it handles the white-glove negotiation as a member of your team.
TractionToday, Vela is already used by professionals with the most complicated scheduling workflows. Enterprise recruiters are using Vela to schedule thousands of multi-party interviews and large sales teams rely on Vela to coordinate cross-functional calls.
Congrats on the launch Gobhanu Sasankar Korisepati and Saatvik Suryajit Korisepati!
https://lnkd.in/g-Aavhsf
You're hunting for talent when your perfect higher just texted So you start coordinating Then another great higher replies Then another and another Another new realized you never responded to the first candidate But what if you didn't have to play that game? Meet Vala your AI scheduling assistant next time your perfect higher reaches out Just CC Vala She'll make sure you get booked She works 24-7 to handle that endless scheduling ping pong So you don't have to Text, Slack, email, Vala understands the context across every channel She manages the noise so you can focus on what matters Building relationships team saved 10 hours a week and they win the best talent Stop losing candidates to your calendar. Let Vala handle it
HookCorvera (YC W26) helps retail brands increase profits by 40% with an AI workforce that handles their day-to-day ops.
They're already saving fast-growing retail brands hundreds of hours a week, helping them serve more customers, and accelerating their growth.
SolutionWith Corvera's platform, brands can:
- Process orders end-to-end without lifting a finger
- Forecast demand in real time
How It Works- Automatically process purchase orders and prevent stockouts
SolutionLed by an ex-CPG founder, Corvera knows exactly how CPG brands work.
Congrats on the launch Christopher Kong, Dirk Breeuwer, and Matthew Collins!
https://lnkd.in/et2HF6Xv
HookRanzo are wasting 40% of their time managing supply chains. When it should be getting to put it out there, it's speaking of customers. Over the last seven years, I saw this firsthand, scaling one of Europe's fastest growing retail brands.
SolutionThey manually process countless sales orders, copying pace from one system to the next and juggle spreadsheet after spreadsheet, or while constantly being paranoid about running out of stock. So we built Corvera, the hands-free command center for fast growing retail brands.
TractionWe hope brands got wasted, improved margins, and maximized revenue with an AI workforce that knows exactly how CPD brands operate. Corvera plugs into existing tools, or to massively process themselves orders and to end, performance to influence.
Corvera knows exactly where your stock is and how much you've got. It forecasts the mind real time, to adjust its corrective action to rent stockouts, so you can maximize your revenue and keep your customers happy.
CTAWe are saving fast growing retail brands hundreds of hours a week, helping them serve more customers and accelerating their growth. Join us, schedule a demo at Corvera.ai. Have a good day.
HookOur healthcare system is fundamentally broken. Eos AI (YC W26) is building autonomous operating systems for healthcare to fix it.
They help hospitals turn their historical operational and clinical data into intelligence they can act on, providing better care at cheaper costs.
TractionEos AI creates harmonized data representations across all systems and time, retrieving similar trajectories to generate actionable forecasts so administrators and physicians can intervene early, save staff hours, and recover revenue. In early deployments, they have seen 3x staff productivity and 37% revenue uplifts.
Congrats on the launch, Arya Khokhar!
https://lnkd.in/gXvM3KMx
HookFor most of history, medicine has waited. We wait for symptoms, we wait for tests, we wait for something to go wrong. But AI promised a new era of medicine.
ProblemWe're carousvastur and decisions are smarter, where we could prevent crises instead of responding to them. Where medicine would become predictive. But here's the reality.
Hospitals have more data than any industry in the world, but only 3% of it is actually used to inform present decisions. Healthcare AI sounds incredible, but it often falls over in the real world, because hospital data lives everywhere.
TractionDifferent systems, different formats, different rules. EOS AI fixes this foundation. We connect the systems hospitals already used to create harmonized data representations and turn their history into a single, searchable timeline.
How It WorksThe platform learns from prior patients in prior workflows and brings that knowledge into the next instance in real time. So AI can work across sites faster and more consistent. The promise of healthcare AI is real.
CloseIt just needs the right infrastructure. At EOS, we're building it.
HookHi, I need some electricity to power my factory supermarket. Data Center. Meet the energy managers.
ContextThey've been assigned to ensure that business can thrive on affordable and reliable energy. Over a thousand-to-locations in different countries. Hello!
How much do you need? Alright, let me see. Can an index contract work for you?
Do you want 100% renewable energy? It can also help you reduce grid dependency with arms very well too. What about demand response?
24-year sites do you want financial hedges and index costs? Are you considering our project? What if you had a friend absorbing that complexity for you?
SolutionIntroducing Condor Energy, your new electricity operating system. Alright, and do you need an index price? Yes, please.
TractionAnd on top of that. I'll have a 20% spot market exposure. I've got some solar panels, so I'll add some storage solution.
And some renewable energy certificates too. Wow, you know your stuff. Thanks, man.
HookBeyond Reach Labs (YC W26) builds deployable solar arrays and radiators that launch compact and expand to football-field scale in orbit.
ProblemCompute in space is growing fast, but rockets are volume-constrained!
The systems that enable next-generation spacecraft — solar panels for power and radiators for heat rejection — only work well when they're large. As satellites, space stations, and in-space data centers demand more power, they need more surface area.
How It WorksBut today's deployable structures don't scale well: they take up too much room when stowed, they're expensive, and they often end up too flexible once deployed. Power generation and heat dissipation are becoming the bottlenecks for the next generation of space infrastructure.
TractionBeyond Reach Labs designs and manufactures patented deployable structures that pack tightly for launch and expand into extremely large, stiff arrays in orbit. Their technology improves stowage efficiency and structural performance, enabling 100kW+ systems needed for orbital data centers and other large space infrastructure.
TeamThe team combines deep flight experience and advanced deployable structures research. Pele Collins spent seven years at SpaceX leading parachute engineering and production across 30+ Dragon missions, and Dr. Mitchell Fogelson earned his PhD at Carnegie Mellon, focused on large deployable space structures with NASA.
They've been building together since 2013 as freshmen in college and are focused on making large space structures reliable, scalable, and practical!
Congrats on the launch, Pele Collins and Mitchell Fogelson, PhD!
https://lnkd.in/g4SBNA6i
HookThis is a G-Type main sequence yellowed wolf, and this was its first friend. Then came this, and this. It needs to be exactly 93 million miles away to harvest life.
ProblemBut now, we're entering a new era. The hardware is here, tools that harness heat and light, solar power that finally works. Data centers at unprecedented scale.
Medicines once the order was brought by energy from a barbe. The power was like no, this is us. And it's finally within reach.
HookCCTV systems were built to record footage. Lexius (YC W26) is building the system that understands it.
When theft or a safety incident occurs, teams scrub through hours of footage to reconstruct what happened. Traditional camera systems store video. They don't think.
How It WorksLexius is the AI layer on top of your existing cameras. They detect shoplifting in real time, alert staff to slip-and-fall risk, and trace individuals across cameras automatically. They do this without any new hardware.
The future of security isn't more footage. It's understanding.
Congrats on the launch, David Elskamp and Liam Webster!
HookCan we enable AI to learn complex physical skills like folding an origami? We believe we can do it by designing a texturist, general and infectory that truly scales with human data. Hi, my name is Daniel.
SolutionHi, my name is Ryan. So together, we're building an origami robotys to make it either the hour for AI to learn physical skills. By optimizing our data collection process and motor design, we can capture and replay fine-grain, hand manipulation data with high fidelity.
TractionWe minimize the existing gaps in finger kinematics, contacts, camera positions, and even friction inside the gearbox to give the AI highest quality of data. We are already working with multiple physical AI labs and some of the companies to deploy our system and push general manipulation further together.
SolutionWe believe the filter of the AI isn't just digital but also physical. So hey, well, that's for some origami and the filter together.
HookAsimov (YC W26) pays households and businesses worldwide to record their daily routines, generating thousands of hours a day of diverse human movement data to train humanoid robots.
SolutionHumanoid robots need to learn from how humans move and interact in diverse environments, not just factory floors. But high-quality movement data from homes, restaurants, offices, and everyday life doesn't exist at scale. Asimov is changing that by empowering people around the world to record what they already do and get paid for it.
TractionAsimov has built a global network of thousands of households and businesses, along with the full stack, from proprietary collection hardware to annotation tooling. They are already providing data to the largest robotics companies in the world.
Congrats on the launch, Lyem Ningthou and Anshul Verma!
https://lnkd.in/ggCMevph
HookEvery movement has a language. But most of us just never learn how to read it. Image models are trained on billions of photos.
For robots, they're still learning how to pick up a glass. Just recording repetitive passing batteries won't cut either. You will always need to handle real-headed vessels.
TractionThat's where we come in. Building the dataset of human movement, first person reporting of real fast, cleaning, cooking, organizing, capture today across thousands of households and businesses. That's part is?
CloseYou can throw it immediately. Turn your daily routine into a living at triasmal.au.
HookSparkles (YC W26) is like Lovable for existing projects, so even your hypothetical Sharon from Operations can change the button styling on the front page without going back and forth with the dev team.
Onboarding non-technical teammates onto Claude Code/Cursor/Codex usually turns into instant chaos.
How It WorksYou've probably heard some version of: "I think I pushed the .env to GitHub… is that bad?", "Terminal is asking for my password but nothing is showing—did my keyboard break?", "It says run git push --force… is that like a Star Wars thing?", or "Do I run npm install every time I open my laptop?"
The CLI + dev server world just isn't a friendly environment if you didn't grow up in it — and it takes real time to learn.
SolutionSparkles is a web platform that fully abstracts local development. Your GTM /Design/Marketing teammates don't need to know what Git is, what a CLI is, or how to run a dev server. They just get a Lovable-like UI with a preview + chat box, and can ship updates as fast as they can type.
TractionWhen they're happy, they click "Upload," and Sparkles turns the changes into a GitHub PR for an engineer to review. Sparkles is also building enterprise rulesets (PR creation policies, commit batching, etc.) so you don't end up reviewing a thousand tiny PRs as your team scales.
Congrats on the launch Ai Daniil Bekirov!
https://lnkd.in/gd5t7H39
Music Meet Sparkles, live coding IDE for non-tiny people. So now everyone can be a builder without breaking promise. Music
HookBeacon Health (YC W26) is building AI agents for primary care - starting with value-based care.
ProblemPrimary care is the primary source of care for over 200M Americans, but physicians don't have the bandwidth to manage everything their patient panels require: preventive screenings, prior authorizations, referrals, and risk adjustment - all critical work that pulls time away from patients.
How It WorksBeacon Health's AI agents handle these workflows directly inside the EHR. They own them end-to-end, putting value-based care on autopilot so clinicians can focus on care.
TractionValue-based care is the $1T market no one knows anything about. Beacon Health is automating it so clinicians can focus on patients - and enjoy practicing medicine again.
Interested? Email mark@beaconhealth.ai
Congrats on the launch Mark Pothen and Obinna A.!
https://lnkd.in/gU5NTsk5
HookPrimary care used to be about people. Morning, Doc. Good morning.
ProblemBut somehow we forgot that. No, he's busy right now. What if?
We could change that to, I think. Hi. How can I help?
SolutionBeacon Health puts primary care on autopilot, starting with value-based care. Our AI agents handle any task in the EHR just like a human, whether it's preventive screenings, prior authorizations, or risk adjustments.
How It WorksWe own these workflows from end-to-end, working behind the scenes, so that doctors can focus on what matters most. They're patients. We're bringing the joy back to medicine, one workflow at a time.
HookCaretta (YC W26) offers realtime AI that enables salespeople to handle the toughest technical conversations with pure confidence.
Sales teams are often responsible for some of the deepest and technical products, and multiple of them at the same time. It's impossible for reps to know every detail, so they fail answering in-call questions, stall deals, and depend on fellow Sales Engineers.
SolutionWith Caretta, reps join calls with a realtime AI that thinks alongside them and enables reps to handle deep product questions, support with qualification and discovery, and more.
How It WorksToday, Caretta helps reps that are responsible for 1000+ APIs, deeptech, enterprise SaaS product suites and many other complex industries.
SolutionHave Caretta join your team at www.caretta.so
Congrats on the launch Kayra Bahadır, Pavlos Markesinis, and Omar Elamin!
https://lnkd.in/darn7QaJ
HookIs that a few simple questions? How do you support encrypted two-way integration, HubSpot? How do you ensure ten-a-isolation?
ContextAlso, is all or something to see? It's a question. Why?
Can you answer that? Yes, I can answer that. With Caretta, who A.E.'s best friend.
SolutionCaretta learns your products and helps your A.E.'s handle complex topics in real time. Okay. Our info sector team is pretty strict with compliant.
And data residency in E. Caretta lives in Slack. You don't have to open any E.
One being in Caretta is in the conversation. Ask it anything. It has the answers.
CloseFire. Make yourselves people technical with Caretta.
HookCumulus Labs (YC W26) is building a performant GPU cloud that preemptively optimizes training and inference workloads.
ProblemAI teams are bleeding money and time on GPU infrastructure. Manual optimization is a losing battle. Teams waste countless hours predicting compute needs, managing failures, and trying to keep utilization high. Meanwhile, clusters sit idle 40-60% of the time.
How It WorksCumulus capitalizes on wasted capacity. The platform aggregates idle GPU infrastructure globally into a single, unified pool. For training and fine-tuning, workloads are predictively packed alongside other jobs to maximize efficiency and dynamically migrated live during execution to faster or cheaper clusters as they become available. For inference, the platform enables ultra-fast cold starts and performant serving where users are. Jobs are charged by the physical GPU fraction consumed, so teams never pay for idle capacity.
SolutionAI scheduling agents constantly diagnose failures, auto-recover workloads, and optimize performance across the entire pool. The Cumulus Prediction system learns usage patterns to optimize resource allocation, ensuring the best performance and price. Getting started takes less than 20 lines of config.
Suryaa engineered custom GPU compute solutions at TensorDock and later built critical infrastructure at Palantir. Veer led a Space Force program and worked on ML workloads and infrastructure at an aerospace startup supporting NASA missions. Together, they're building the solution the industry needs.
Congrats on the launch Veer Shah and Suryaa Rajinikanth!
CTAJoin The Waitlist: https://cumuluslabs.io/
https://lnkd.in/enHYrxrE
HookThe GPU shorter GPU shorting The GPU shorting is for the coming year. Modern Autofishing Intelligent AI Powered Website Hi, valuation of AI TechFurbs Every year, billions of dollars are being spent on GPU compute And yet, most teams still can't get the right GPU at the right time for the right price.
SolutionIntroducing cumulus, a platform that migrates your workflow's live, shares V-RAM globally and only bills for resources used, not idle time. Traditionally, customers are locked out of the latest far-where, and are forced to pay in St.Prices just to babysit their jobs.
With cumulus, you submit your job months, and you let us handle the rest. For training and fine tuning, we intelligently pack you on to our clusters using the latest hardware to enable the most performant jobs.
TractionWe aggregate trusted providers from around the globe to ensure high availability, low latency, and the cheapest rates on the market. For inference jobs, we capture your live execution states and replicate them across our multi-tending clusters, letting you scale from one to a billion where your customers actually are.
Our platform continuously migrates, schedules, and optimizes jobs during its front time to save 50-70% of all costs. And when your job hits a snack, we automatically diagnose errors for a great stop time.
CloseAccumulus, we're building the GPU cloud on the future, and we can't wait to power your business.
HookCardinal is the AI platform for precision outbound. They already power outbound for 40+ companies like Greptile, Luminai, Mintlify, etc.
"Spray and pray" is dead - the teams winning outbound do it with extreme precision: finding the perfect leads, monitoring their content, tracking signals, qualifying inbound, generating hyper-personalized snippets based on commonalities, surrounding your ICP, etc.
SolutionBefore Cardinal, that meant 10+ tools and a GTM engineering team. Now, Cardinal's agents handle all of it.
Founded by Devi (CEO) and Jianna (CTO), 2x YC founders and MIT and Harvard CS alums.
HookOptical satellites fail about 70% of the time, yet defense strategies and market decisions depend on visibility that can be blocked by clouds and darkness. Modern intelligence demands reliable ground truth in any weather.
AxionOrbital Space (YC W26) has built advanced SAR-to-Optical foundation models on Earth to solve this problem.
SolutionTheir Deterministic One-Step architecture resets the SOTA with an FID of 30.24 and SSIM of 0.6 by anchoring generation to physical spatial priors that strictly constrain outputs to reality. Their models, Hubble (10m resolution & Open Source) and Orion (0.5m & Restricted), achieve 0.06s inference latency for live monitoring.
From air-gapped servers to the satellite edge, you can now track any asset through day, night, or storm.
Congrats on the launch, Dhenenjay Yadav & Atharva Peshkar!
https://lnkd.in/g6ypdKnM
HookAnybody there? For years, we've used satellites to image the Earth. We've built a world that can be seen from orbit.
ProblemCopy that, but the Earth doesn't always cooperate. Clouds roll in, after planet goes dark, but markets don't wait for daylight or clear skies. Radar sees differently.
It cuts through clouds. It works at night. And with our foundational models, we translate radar into clear, beautiful imagery.
Track oil reserves, any hour. Track ships in real time. Few enemies on any given night.
Hooknoetic (YC W26) is making hardware compliance 10x faster.
ProblemGetting hardware to market means navigating thousands of pages of fragmented standards, expensive consultants, and opaque back-and-forth before a single test is run. One mistake can mean months of delays, re-testing, and lost deals.
How It WorksNoetic automates compliance workflows using AI agents that understand the standards, build the technical files, and route products to the right testing labs. Industry professionals handle final review before anything ships. Teams hand over their specs and get a certified product back in weeks with no paperwork, no engineering hours, and no consultant calls required.
The founders are Yale dropouts with roots in frontier robotics, from perception research to humanoid deployments.
Congrats on the launch, Tony Gao, Jack Yin, and Henry Zheng!
https://lnkd.in/gfmmx3af
HookThe way hardware companies navigate product approvals today is broken. Before a product ever reaches a customer, there's a wall of obstacles most founders underestimate. Compliance.
ProblemToday, hardware compliance is slow, fragmented, and incredibly expensive. There's no simple equivalent of SOX2 for hardware. Instead, there are hundreds of different standards depending on your industry, the country you want to sell in, and the requirements of your customers.
The information technically exists, but accessing it is a whole different story. Companies will spend tens of thousands of dollars on lawyers and consultants just to understand what regulations apply to that, but it doesn't stop there.
How It WorksMatt testing can take up to several months, and if the product doesn't pass certification, teams are forced to redesign, losing both time and capital. Hardware is already hard enough to build. How many think about compliance, too, is just overkill.
SolutionWhich is why today, we're launching new edic. We've learned AI agents and licensed professionals to navigate the entire process, so that you don't have to think about standards or paperwork at all. Our research agent analyzes your product specification and target markets.
TractionIt then parses through thousands of regulatory documentation, and identifies exactly what standards matter to you and what you need to understand. Next, our documentation agent drafts, technical files, and test plans required for certification, helping you prepare for testing so that you can talk to customers.
But we don't just stop there. We leverage our global network to match companies with the right testing labs for their specific product. The result is a process that is faster, more organized, and far more affordable than traditional consulting.
CloseNo edic makes compliance seamless, so hardware companies can do what they do best.
HookLibrar Labs (YC W26) Labs have built the AI librarian that saves library staff hundreds of hours and inspires kids to read.
SolutionStudent reading rates have hit an all time low. That's why Librar Labs built a library system and an AI librarian that can do the busywork saving library staff hundreds of hours and more than doubling reading rates.
Librar is built on a unique, self-healing database infrastructure that is transforming how AI can navigate the digital world of literature. This powerful technology is now used to more than double reading rates for the hundreds of paying schools of Librar, and is being adopted at a pace never seen before in publishing, library and archive tech history.
TractionThe team of quantum physicists, exited founders and dropouts are backed by founders and operators from OpenAI, Depict, Kahoot and many more, to tackle the one of the greatest problems of our time.
librarlabs.com
Congrats on the launch, Jonathan Görtz, Carl-Hugo Jacobsson, and Kaan Sirin!
https://lnkd.in/guNa2FND
HookAll over the world, school libraries are failing students' reading needs. In the US, three are four schools operate with alcohol if I laboratory them. Running a library is already really hard, and without a librarian it's basically impossible.
The library labs is a library system and a library that makes it a lot easier to manage the library. With a library can be set up by taking images of the shows. Our system identifies the collection and organizes it.
Library provides insights into inventory, acquisitions, and usage, saving librarians hundreds of hours. We have the ultimate goal of eliminating illiteracy and where I'm trying to do so. In using libraries, some schools have double the reading rates.
CloseIt's time to make libraries more than just a room on books. It's time to put books back as students have.
HookSeeing Systems (YC W26) is building inexpensive autonomous strike drones designed to stay operationally relevant as warfare evolves. By combining true hardware modularity with an agentic control system that lowers cognitive load on operators, they enable faster upgrades, broader adoption, and more than 2x reduction in lifecycle cost.
ProblemCustomers and partners include the Royal Marine Commandos, several other units in the UK, and 4 other NATO forces. They are currently shipping prototypes to Ukraine for iterative feedback.
How It WorksSeeing Systems maintains tight innovation loops by embedding forward-deployed engineers within units, integrating real-world feedback into systems in days, not months.
TeamBuilt by brothers Alex and Matt Le Maitre, they combine previous defense experience across hardware and software with a focus on building the systems our operators actually need.
Congrats on the launch Alexander Le Maitre and Matthew Le Maitre!
https://lnkd.in/dV6dtvsc
HookHi, I'm Matt and I'm his brother Alex. No, we're not another AIB to be sass. This is seeing systems, a defense startup building modular strike and reconnaissance system that require no operator.
ContextI'm talking ungeneral autonomous drone dominance. Six months in, we've already collaborated with a number of armed forces across Europe and one thing is clear, these little things have redefined the modern battlefield.
TractionYet, these cutting-edge drones are still basically expensive toys. Ukraine uses at least 200,000 of these every month with only 5% of the achieving their mission. Often thousands cannot be used due to EW, that's jamming.
SolutionWe've lost friends in Ukraine, so we understand how important these problems are. And that's why we're building something, just as affordable, twice as effective and 100 times easier to use. That's why we're building Banshee.
This is a biologic medication and this tiny viral it costs 30 thousand dollars Sorry, well pay for this And that's $30,000 gone just like that this is how in fusion clinics feel when they're battling insurance for prior Authorizations the office staff submits a prior off before the patient comes in for their infusion But sometimes insurance doesn't respond in time Patients still need their medication so clinics will still treat them but them boom denial insurance companies won't cover this infusion I made prior authorization coordinator here at Sanctuary Metology and even after all this the prior often still get denied This delays patient care and her third clinic's financials. We knew there had to be a better way to do this That's why we created room a care We hope clinics get patients other special to medication faster First we show staff exactly what they need to get a prior off the approval Second we've sold for data entry mistakes through automated form filling And third supporting patients on the financial side with coping enrollment assistance Well, this is pretty good, but I'm still docking is a 30 thousand dollars from your pay
HookWe are giving Consciousness to AI, which means AI will start behaving like humans. What's that actually mean? Right now, there is an AI out there which has social awareness.
ContextFor example, Chagipity Voice is socially dumb. If you're talking with your friend and Chagipity Voice is open, it will always assume you're talking to it. That's wonderful, I'm it.
How would I lead a ride? Shut up! We are creating fun.
CTAAn AI that will converse like a human. For her nose went to say silent and went to be proactive. Check it out.
I think we should totally switch. Not sure man, right now it's pretty good for me and notes. Hmm.
Any ideas for her? Yes. If you just want notes after the fact, then yeah, we can stay on course.
ProblemBut if you want an AI teammate that can help in your meetings in real time, then I think switching is the better option. Alright, can you draw a female to cancel the subsequent please? No problem.
CTAAlright, check it out. That's good fun. Thanks.
How It WorksWith the app, you can seamlessly choose which meetings firm will join and get all the insights about the meeting. In the future, integration of fun with smart glasses, robots and more will make all of them socially time and contextual.
CTAIn other words, we will make these tools human. Try for an in your team discussions today. And if you're looking to integrate socially aware AI into your products today, feel free to reach out.
ProblemTool calls error or time out. Agents claim they "did it" without actually doing it. Prompt injections leak data. And long conversations make it hard to see what actually broke.
SolutionModa (YC W26) is monitoring and reliability for AI agents.
It turns conversations and tool traces into actionable signals; detecting failures, hallucinations, prompt injections, and regressions so teams can quickly see what's actually breaking in production.
TractionIf you're running AI agents, Moda helps you catch issues before users do.
Congrats on the launch, Mohammed Al-Rasheed and Pranav Bedi!
https://lnkd.in/gfvYbjpr
HookWhy did your open-cloud agent just promise our users full refunds? Refunds too, are you kidding me? I'll get on that next.
ContextWhat do you mean, too? What's going on? You wanted more AI, so I added it.
ProblemBut there I can't go through a million logs. There's no easy way to get alerts when it messes up. Oh, baby, what?
Right now, I can't allow for my kid. I can't get in. Don't worry the time.
If I'll make your hand full, no my shoes. I'm a fan to me. I can't stay in it.
I'll be back. It's okay. I'll get to see you.
No, no, no, no. We can't try. What does it mean?
You're the best I've been for me. When you're the brother of Westfield, I'm not the only one who's taking the love. Stay in the love.
We'll see you bring it back to body. She can't stay in the love. Stay in the love.
HookFort (YC W26) is a wearable that automatically tracks strength training for people who care about longevity.
ProblemStrength training has become a core part of how people think about health and longevity. It is no longer a niche activity for gym enthusiasts. More people are lifting, doing Pilates, eating protein, and watching fitness content than ever before, yet the wearables they use still focus almost entirely on sleep and cardio. Despite the shift in behavior, strength remains poorly measured and poorly understood by today's devices.
How It WorksIf you want to track strength workouts using existing wearables, you are usually forced to manually log exercises, sets, and reps. Some platforms attempt automatic rep or exercise detection, but the experience is unreliable and far less polished than cardio tracking. Even when workouts are logged correctly, wearables largely ignore muscular fatigue and provide little guidance on how to train, recover, or progress toward meaningful goals.
SolutionFort uses motion and heart rate data to detect exercises, reps, sets, movement quality, and fatigue without manual logging. The device is a small, screenless band that can be worn on the wrist or attached magnetically to gym equipment to capture movements that wrist-only wearables miss. Fort can be worn all day, lasts about a week on a single charge, and also tracks steps, sleep, heart rate, and cardio to give a complete picture of health.
TractionIf you're interested in trying Fort, pre-orders are available at fort.cx/order. A limited quantity of early-access beta test units are already shipping, and production units will start shipping in June 2026.
Congrats on the launch Miranda Nover, Zac V., and Paul Schneider!
https://lnkd.in/gXDzVKDW
HookFort is aware of all that automatically tracks strings training for people who care about longevity. The data is clear. Strength is crucial for health and more popular than ever.
ProblemHowever, tracking strings training requires you to manually log every rep set and exercise in the gym. And even then, the analytics you get on your strength are less accurate, less useful, less actionable, than what you get for cardiovascular sleep.
How It WorksWe're fixing that. Fort automatically attracts your reps, sets, and the exercises you've performed. We've provided feedback on your form and training intensity to help you reach your goal faster.
CloseThe device is designed to be worn all day and track your sleep, steps, and heart rate like you're used to. We've just opened our pre-order and you could purchase a device via our website at Fort.cxslashorder.
HookEnd Close is building automatic reconciliation for payments companies.
TeamCEO Sean spent 6 years building the reconciliation org at Modern Treasury, processing a trillion dollars in payments across 40 banks with 99.995% automation. CTO David is a second time YC founder and previously led error tracking at PostHog.
How It WorksEvery fast-growing fintech, marketplace, and payroll company hits the same wall: reconciliation breaks. Unlike legacy tools built for month-end close, End Close reconciles continuously with developer-friendly APIs so teams can power real-time product experiences like instant balance updates and faster trading.
Every exception used to mean hours of manual investigation. End Close's AI agent connects to your data warehouse, processors, and invoicing platform - and resolves them automatically.
With new payment rails and agent-to-agent payments driving higher transaction volumes every year, reconciliation is only getting harder. End Close makes it automatic.
Congrats on the launch, Sean Bolton and David Newell!
https://lnkd.in/g82ZG2cG
HookKeeping track of money is not as simple as you might think. Payment companies move money through banks, payment processors, ledgers and data warehouses, and oftentimes things don't match up leading to money loss.
SolutionOn reliable financials and compliance risks, legacy solutions are designed around month-end clothes, which doesn't work when accurate real-time data is needed for features like instant deposits, and nothing solves the hardest part, exception handling, which is where our team still spend hours every day. This is why today, we're launching end-clothes, a continuous reconciliation platform that increases automation over time with AI agents that resolve exceptions.
How It WorksI spent the last six years automating reconciliation of a trillion dollars, and with end-clothes, we're bringing this to every company. End-clothes works with their existing staff. Connect your services or create records through our API.
To reconcile any combination of data's principle, and when something doesn't line up, our AI agent steps in. It pulls context from your connected systems automatic. It identifies the root cause of the issue and results without involving your team.
We're used to take hours, now takes a few minutes. Happens autonomously, and is entirely transparent. End-clothes closes the loop for modern payments companies.
CloseSo you can stop worrying about where the money went, and start building what's next. End-clothes.
HookWe like to think our world is solid, built to last, but look closer. It's a house of cards waiting to fall. Every click, every login, every transaction.
SolutionIt's fragile, power grids, hospitals, banks, built for speed, not built for defense. And someone is always watching, probing, waiting for the weakest link. You run audits and call it a secure, but they're watching the cards, not the players.
So we build hex, AI agents that move first, testing your security with algorithmic speed and surgical precision. Injection flows, access gaps, blind spots, hex finds them. The system isn't changing, but now you move first.
HookOver 20 years in aircraft maintenance of seeing firsthand technicians frustrated at spending too much time on paperwork and not enough time on the aircraft. That's why we built simply the AI co-pilot that automates technician admin.
ProblemFinding is raised in seconds simply connects the thoughts between manuals, local procedures, and trial and knowledge across systems. The parts are needed, simply finds a right part number, telephone effectiveness, inventory, and interchangeability.
TractionOnce the work is done, the technician speaks, is simply transcribes and drafts compliant notes. Every work order gets a second pair of eyes. Zimdly audit actions and flags issues before they cost you millions.
DemoThe technician's shouldn't have to lose half their data admin. I'll just give you a team time back. Okay demo with Zimdly.
HookCellType (YC W26) is the agentic drug company: AI agents + foundation models that simulate human biology to discover drugs.
How It Works90% of drugs fail because preclinical models don't match humans. CellType's foundation models simulate what happens in patients before the trial, while AI agents run the full discovery pipeline end-to-end.
TeamTheir core technology, Cell2Sentence, was developed with Google DeepMind and shared by Google CEO Sundar Pichai. They've already used it to discover and validate a novel cancer treatment signal. They're working with Top 10 pharma.
SolutionDavid van Dijk is a Yale Professor (11,000+ citations; Cell, Nature, ICML) who turned down Google to build CellType. Ivan V. co-developed the core technology while at Yale (published at ICML), led foundation model training at another biotech startup, and built software to control CERN's Large Hadron Collider.
Congrats on the launch, David & Ivan!
https://lnkd.in/gbSDZPkA
HookDrug development is broken. It takes over 10 years and billions of dollars to bring a single drug to market. And even then 90% fail in clinical trials.
ContextMost drugs don't fail because the chemistry is wrong. They fail because they're testing in the wrong models. My since-sell lines don't reflect what actually happens in humans.
SolutionWe started self-dip to fix this. The first agente drug company. We built foundation models that learn human biology, not just proteins, but the whole system sells my environments, tissues, and how they respond to drugs.
TractionThink of it as a virtual human. You can test the drug on it and predict what will happen in patients before you ever run a trial. We already used it to discover new cancer treatments.
TeamA drug that can help turn cold tumors hot. Our research was featured on the Google blog and shared why Google CEO soon their future. I'm David.
CloseI'm your mom. We are Cell Type, the agente drug company.
It's a new dawn, it's a new day, it's a new life for me And I'm feeling good I'm feeling good It's a new dawn, it's a new day River running free, you know how I feel Blast them on the tree, you know how I feel It's a new dawn, it's a new day, it's a new life for me And I'm feeling good
HookSponge (YC W26) is the easiest way for agents to hold and spend money, and for businesses to sell directly to them.
TeamBuilt by the team behind Stripe's stablecoin financial accounts, stablecoin payments and core money movement systems, Sponge brings deep experience in building financial systems for the new economic actor: agents
SolutionFor agents: Sponge lets agents hold funds and transact autonomously with other agents and businesses using bank accounts, cards, and crypto. They can pay, invest, and earn money without a human in the loop.
How It WorksFor businesses: Sponge makes it easy to onboard and accept payments from agents without human interaction. Businesses can sell their services directly to agents with a few clicks, no code changes required.
SolutionTeams like AgentMail and Steel are using Sponge to onboard agents and receive payments from them.
CTATo give your agent an account or to sell to agents, visit paysponge.com or contact team@paysponge.com.
Congratulations on the launch Eric Zhang, Jae Choi, Rishab Luthra!
ProblemCompanies spend $1T on ads every year. $100B is wasted on ineffective ads due to non-compliance with brand guidelines, regulations, and platform best practices. So millions of lawyers, marketers, and compliance officers manually review every piece of content.
This process is slow, expensive, and about to break - AI tools are letting teams create 100x more content and human reviewers won't be able to keep up.
TractionVeriad automates the entire process at scale. It ingests brand policies, industry guidelines, and precedents, then analyzes every piece of content automatically, providing instant feedback to marketers. Veriad integrates directly into your marketing stack, so all of your content is on-brand and compliant 100% of the time.
Congrats on the launch Rohan Mahendraker and Anton Muratov!
https://lnkd.in/eHW7h8kz
HookYou are about to make a million dollars working on your masterpiece ad campaign. There's just one problem. This guy, Mr Reality Check.
Revisions, approvals, email chains. Your big ideas stuck in endless review, so you just give up and he wins. Don't let compliance stop your campaigns.
How It WorksVery ad ingest your brand guidelines, regulatory codes and best practices. An analyzes all of your marketing content automatically. Telling you what's compliant, what's not and why.
ClosePublish faster and stay compliant with very ad.
HookPiris Labs (YC W26) provides a full-stack, high-speed inference solution that significantly reduces cost per token without compromising model accuracy or performance.
SolutionThe breakthrough is based on optical technology that removes the communication bottlenecks within and between compute and memory, coupled with a purpose-built software stack. Today, the "memory wall" is forcing the industry toward band-aid solutions, increasingly complex and expensive memory packaging, making it the single greatest barrier to profitable AI.
How It WorksThrough vertically optimized optical interconnects and a dedicated software stack, Piris Labs is reimagining data center architecture. They enable cost-effective memory and compute to operate seamlessly at any distance while delivering 5x lower latency, 10x lower power per bit, and 2x lower cost per token.
SolutionBy transforming the data center into a single, coherent compute node, Piris Labs enables inference at the speed of light.
TractionIf you want to scale your inference service while slashing costs, reach out at contact@pirislabs.io. Piris Labs is also looking for additional partners who get the chance of trying their optical solution first hand. Shoot them an email at founders@pirislabs.io.
Congrats on the launch Ali Khalatpour and Keyvan Moghadam!
https://lnkd.in/gUpkwmRN
HookAI is the most transformative technology of my life. Low AI is expensive. In fact, most of the AI companies now are losing money.
ProblemThe problem is not just software. It's infrastructure. Data movement is emerging as the major bottleneck and the primary driver of the cost.
TeamIt's low and power friendly interconnects as cause the memory wall, which it responsible not only for the high cost per token, but also the significant increase in time to the first one. And K-1. I met co-founder at Pirates Labs.
I'm an AI scientist and engineering leader. I'm an optical scientist. I have a specialized in building and developing into enterprises.
How It WorksAnd we came together to deliver the best end to end solution here at Pirates Labs. We have replaced the power hungry and a slow copper interconnects with our unique optical interconnect. The result can actually increase the power consumption per bit and also find that increase in bandwidth per length.
To deliver the best, we have a couple of tests that we've heard the still software stuff. Our food worked with integrated solution results in our explore and latency and significantly reduction in cost-free.
TractionIn less than three months, we have to build our own characterization life. We have been building our own for chronic chance, and we have been able to develop the most efficient conversion engine content.
We are now working with the variety of companies interested in accelerated inference. We have been also working very closely with the people at the cost, which is now part of the building. So if you need better inference service and like to have early access, please contact us at contact at Pirates Labs.io.
HookCanary (YC W26) is the AI QA engineer that understands your codebase. AI coding tools have broken QA. Customer-facing incidents are up 43% year over year.
SolutionTeams are shipping 5–10x more code. PRs are bigger and code reviews still happen in file diffs, not in real user journeys. That's why changes that look clean can still break checkout, auth, or billing.
How It WorksMost testing tools start from the outside. DOM, recorded clicks, and web agents all try to reverse-engineer what your app is supposed to do.
SolutionCanary starts from the code. It reads your PR to know what changed, uses your codebase to understand how to test it, and runs feature and regression checks against your preview app. If a user flow breaks, you know before production.
TractionTeams are already using Canary to catch broken auth flows, slow page loads, and failing checkouts before users hit them.
CTABuilt by ex-Google and Windsurf engineers who helped build the AI coding tools that accelerated this problem. Now they're building the fix. Check it out at runcanary.ai
Congrats on the launch, Viswesh NG and team!
https://lnkd.in/g6r4uyyY
HookJust one bug, Shaktus, Pras, Belp O'Eos. It's to enter by a contract worth millions that joins, because in its case, cost the data leak. And that's why we will be Canadian.
ContextWe will disobey for AI food. Here is how it works. A developer raises a PR.
SolutionCanary gets to work. It analyzes the dev, figures out which user workflows the future impacts and creates tests for each of them. Not just happy paths.
Every combination a user could try. Video recordings drop right back into the PR. Any bugs get flagged all before murder.
Got something specific to dust, ask Canary to run it. Right from the PR. Lot more coverage, just type what to test.
How It WorksThose tests may become part of your regression suite, automatically. Canary understands your branding logic, generates tests for full coverage, and runs them on every commit and on nightly schedules automatically.
TractionIn the early days of high-pressure, we actually had a customer go in and they ran into all these issues and bugs. And that customer actually ended up not signing the cost. We decided to double down on QA.
We were spending hours chasing down random UIT bugs. Your most valuable and most expensive research in the startup is your engineering's time in bandwidth. It's tens of thousands of dollars.
CloseThose hours that they spend, you can't get the back one say to secure. We are building the well-resigned LIT for AI Code. Check us out at runcanary.ai.
Hooko11 (YC W26) is now used by companies and prosumers all around the world. We're growing extremely fast and are empowering new customers every day.
ProblemToday, we're launching McKinsey-level, editable charts in PowerPoint — straight from natural language. That's right, you can now create and edit 3d visualizations, Gantt, Waterfall, Mekko, and countless other charts with just words. This has NEVER been done before. We're also launching two more features: infinite context, and comprehensive Excel audits (more below).
How It Workso11 is the best tool for automating a M365 workflow. If you work at a company that wants to move faster, book a demo with us at https://o11.ai/book-demo.
CTAIf you're a consumer and want 300 FREE o11 credits, sign up at https://o11.ai, like/repost this post, and comment "AUTOMATE" below.
If you're an investor hoping to invest, we'll be raising our next round in March. Feel free to schedule a time with us at https://lnkd.in/gRdz6S7k.
A bit more about the features we're launching:
SolutionEnterprise-grade PowerPoint charts: Instead of spending hours manually tinkering with think-cell or UpSlide to get your charts to look perfect, just ask o11 to do it with simple words. Every creation and edit is handled seamlessly for you. Instead of gatekeeping such technology behind an enterprise license, we're releasing it free for the masses.
TractionInfinite context: With o11, you can now upload infinite amounts of files without ever having to worry about models exceeding context. Our systems enable o11 to securely process and traverse thousands of your files in a single chat.
SolutionExcel Audit: You can now audit massive workbooks through o11 and receive a list of every spreadsheet error, logical error, and factual inconsistency you made.
TractionThank you to the entire o11 (YC W26) team for all their work, LeapYear , and our Y Combinator partners, Jared Friedman and Francois Chaubard, for all their advice.
HookTwo weeks ago, we reminded the world that Microsoft Co-Pilot sucks. We then launched O11. The best AI productivity tool built for Microsoft 365.
SolutionToday, leading company as a pro-serum as a cross-serve world, use O11Dodemates, spreadsheets, decks, and documents. Here's how. In PowerPoint, teams use O11 to make editable decks directly inside the app.
Everything is templated and synced across Microsoft 365 to enable the best user experience. We're also excited to announce a launch of O11 charts. With O11, you can now create McKinsey level charts directly within your workflow.
Instead of current add-ins, we have to edit things and position things around manually. With O11, it's all done through natural language, which allows you to move faster. With O11Dodemates, so, modeling shifts from taking hours of work to real-time results.
TractionHere, O11 is shown already in a spreadsheet, checking for factual inconsistencies, spreadsheet errors, and formatting errors in seconds. Stuff you might overlook. In word, legal teams, financial firms, and startups are using O11 to draft format and redline documents in seconds.
No more hopping from app to app to copy and paste format. With O11, everything stays in one place. We're also excited to announce a launch of infinite context.
How It WorksWith the infinite context, you can upload unlimited documents to each one of the add-ins. Companies are using this to analyze years of data with an Excel, PowerPoint, and Word. We're the first company to integrate across the Microsoft 365 suite but we're not stopping there.
SolutionSoon, O11 will be the default, available in Microsoft, Google Workspace, and every other productivity tool on the market. Try us for free or book it out here at O11.AI. Thank you.
HookWhat if America and Lady Greenwood would have role never felt what if the Soviet Union survived? What if you could reshape the modern world most strategy games for Sudan pre-written paths? A handful of fixed choices.
SolutionThe same outcomes every time. Pax's story is different. Our AI agents don't follow scripts.
They remember what you've done. They adapt to decisions. They negotiate a train and respond.
We have over 35,000 people playing game. Let's say it's the modern day. You're the United States and you want Greenwood.
Do you negotiate with Denmark? Offer a trade? Do you invade?
Say you try to negotiate. Denmark's AI rejects your offer and calls upon native. The EU responds with sanctions.
Every choice cascades. And every nation of dance. Anyone can build a scenario.
Drug custom borders set any starting year. Write the prompts that shape the AI. Then publish it for others to play.
Our community has published over 4,000 playable worlds. The community builds, players pay and creators earn the tokens. This is Pax's story.
HookStrand AI (YC W26) builds foundation models that predict missing biological data, turning incomplete patient profiles into complete, multimodal datasets, enabling better treatments in pharma drug programs.
Problem9 out of 10 drug trials fail. Often, it's not because the drug doesn't work, but because the right patients weren't selected — the biomarker signal was hiding in a modality that was never measured. For any given patient in a clinical trial, you might have their pathology slides but no proteomics, their genomics but not their RNA expression. The full picture is never there, and collecting it all is expensive, invasive, and often impossible.
SolutionStrand AI trains cross-modal prediction models that take what you already have and predict what you don't: spatial proteomics from H&E slides, gene expression from genotype, and more. Their first foundation model predicts spatial proteomics from standard histology, beating SOTA. Pharma and biotech teams can now screen cohorts for biomarkers they never assayed, stratify patients more precisely, and de-risk clinical trial decisions, all from data they already have on the shelf.
TractionBefore Strand AI, Yue and Oded worked together at Enable Medicine building the platform for petabyte-scale multimodal spatial biology. Yue previously built biology ML infrastructure at scale, working with the founders of Tempus AI. They know what multimodal biological data looks like at scale: what's missing, what's noisy, and how to fill the gaps.
Congrats on the launch, Yue Dai and Oded Falik!
https://lnkd.in/dnwEJpMS
HookImagine if we could see the full biological picture for every patient, every gene, every protein, every signal that matters. Imagine the discoveries we could make, the lives we could save, the possibilities are endless.
And in reality, most patient data is incomplete and fragmented. For nearly all patients, complete data simply does not exist. Every year promising treatments fall short, patients go untreated and billions are lost, we're still operating without the full biological picture.
TractionBut what if we already have everything we need? Biology is deeply interconnected, every modality leaves traces in the others, we just needed to pull on the right strands and predict what was never measured.
How It WorksWe're making connections that were never possible before now and filling gaps once thought impossible. Strand AI is building the models of the future for drug development. And AI, the data layer for biology.
HookCorelayer (YC W26) is the AI on-call engineer for data pipelines.
SolutionWhen you're on call in data-heavy industries like financial services and fintech, healthcare, or insurance, you need to inspect data to debug production issues. Corelayer monitors both data and infrastructure for anomalies and uses AI agents to debug and suggest fixes in minutes.
How It WorksData is especially sensitive in regulated industries, so Corelayer offers on-prem deployments, confidential compute (hardware-backed secure environments), and an audit trail with source-backed citations for each agent action.
TractionCorelayer was founded by Mitch Radhuber and Shipra Jha. They built data infrastructure together at Goldman Sachs, where they spent many late nights and weekends debugging systems that processed 100s of billions of rows a day.
Congrats to the team on the launch!
https://lnkd.in/eTma5TMA
侫 For layer is the AI on call engineer that monitors your logs and data for issues and tells you what went wrong and how to fix it. Its design for teams building data heavy systems where debugging is impossible without inspecting data. Here are data anomaly triggers and agents. It checks logs, code, deployment history, past issues, and upstream data. Find the root cause in minutes and suggest a fix.
HookOpalite Health (YC W26) is an AI medical interpreter that helps healthcare providers communicate with patients who don't speak English.
Healthcare's biggest failure is also its most basic: making sure doctors and patients can actually understand each other.
SolutionCathleen K. started Opalite after watching her immigrant parents struggle in the US healthcare system - simply because they don't speak English.
When a human interpreter mistranslated her mom's symptoms, it led to serious complications and a return trip to the hospital.
TractionLater, she saw the same problems from the physician's perspective. Her hospital spent millions per year on interpretation services, yet providers still wasted hours waiting for interpreters who sometimes never came.
She knew something had to change.
Opalite is available instantly, 24/7, integrates directly into the EHR, and replaces traditional interpretation services with software that's faster, more reliable, and over 50% cheaper. It's already live across hospitals and clinics, translating 10,000+ minutes per month.
Congrats on the launch Cathleen K. and Alex Mehregan!
https://lnkd.in/eVAVPtrg
HookThis is going to change healthcare accessibility forever. For over 30 million non-kinders speaking patients in the United States, healthcare communication is broken. I wouldn't know because I'm an immigrant and a physician.
ProblemI saw this firsthand and an emergency department. I had a patient crying in pain and I couldn't understand what was wrong. We tried calling a phone interpreter, but it took forever to get someone online.
We tried Google Translate, but it didn't make sense. And even after my hospital pays millions of interpretation. It still breaks down when patient needed a moose.
SolutionHow can you be something purpose-built? Introducing Opelate. Opelate is an AI medical interpretation app for healthcare providers.
How It WorksOur AI model is trained on extensive medical knowledge, making it translations far more accurate than anything else on the market. We also integrate into every popular EHR, making documentation a breeze.
TractionWith Opelate providers don't have to slow down, patients finally feel understood, and higher quality care is possible. We're already live in hospitals and rolling out in large enterprises, or translating tens of thousands of minutes per month and are trusted even in the most high-stakes situations.
CTAOpelate makes treating non-English speaking patients as easy as anyone else. Join us to make healthcare more accessible to all patients. Regardless of their language or background, thank you.
HookLaurence (YC W26) automates performance marketing for CPG brands, starting on Amazon.
ProblemToday, brands outsource advertising campaign management to agencies and offshore teams who manually set bids, budgets, and keyword targets using copy-paste spreadsheets, barely once a week.
These simplistic rules ignore inherent randomness in consumer behavior, and lousy performance causes most brands to lose money advertising after margin.
SolutionLaurence uses reinforcement learning to optimally set bids, budgets, and keyword targets like quants trade markets, adjusting every single hour.
TractionThey are automating $10 million in live ad spend, increasing advertising profits by 15-20% while growing total brand revenue.
Congrats on the launch Matthew Chen and Leo Gierhake!
https://lnkd.in/g6KhSu4V
HookHey, I'm Matt, I'm Leo, and we just quit our jobs in Big Tech and Quantfinance to build Lawrence, the future of performance marketing. Back by white culminator Tom Aynhozen's scout funds were setting out to save CBG France on Amazon Billions, a year in Wasted Acet and Los sales.
ProblemOur fund system currently automates $10 million in live asset. Weasers customers existing at Colby, optimizing bids and budgets across tens of thousands of keywords of dating every single hour. As a sather, you said bids on keywords ahead of time.
Then when someone serves on Amazon, a real-time auctioneer stream. Today, teams and aid runes in manually address bits for only a few keywords, barely once a vote. Overpaying $10 per claim.
TractionNow that doesn't sound so bad, who on each keyword gets 10 clicks a day across 50 keywords per product. A small brand that sells for 50 skis is losing almost a million dollars every single year. We use reinforced learning and trainer-owned LLM so that Lawrence can let your Amazon ads on on-of-ad.
Harvesting profits will be confident and borrowing signals for similar keywords with data as far as. Unlike black box agencies and bureaucratic in-house teams, Lawrence is a fully transparent always on system.
CloseWith our Ask Laureates AI agent, you can understand exactly what drives your performance whenever you want. Lawrence is a quantitative brain for performance marketing. To work with us, email us at MatthewEntryWood.com.
HookOpenSpec (YC W26) is building the open-source spec framework for coding agents.
ProblemGenerating code is now cheap. Correctness is still expensive. Engineers don't distrust AI-generated code because models aren't capable. They distrust it because they can't reliably get the output they actually wanted. The bottleneck isn't the model. It's underspecification.
Teams that have adopted AI coding tools are feeling this everywhere. Code review feels heavier. Agents produce work that doesn't quite fit. Codebases are changing faster than anyone can track. The answer isn't better models. It's giving developers a way to generate living specifications that evolve with their codebase, making their agents reliable enough to take on larger, more complex work.
TractionOpenSpec has grown to 28k+ GitHub stars in under six months, with thousands of developers and hundreds of teams using it in production. Try it at openspec.dev
Congrats on the launch, Tabish Bidiwale!
https://lnkd.in/gZFAM3v6
HookWhat would it take for a coding agent to be completely self-sufficient? To build out a feature, end to end, completely autonomously. Models have gotten remarkably capable.
ProblemOn paper, it feels like we're almost there. But developers still struggle to use agents reliably for complex production work. To understand why, let's start simple.
Agents make assumptions. Each assumption is a branch. When scope is small, it's easy to catch and correct.
But branches compound as the work gets bigger. And this isn't an intelligence problem. The path might not be wrong.
Our mental model is just different. The agent is guessing at our intent because our intent was never specified. We need something more structured than plan mode.
Something persistent. An artifact for coordinating an agent through assumptions, decisions, constraints, and requirements. That's what a spec is.
And open spec is the open source framework that puts it into practice. You structure your intent, go back and forth with the agent, surface assumptions, catch wrong turns before they become wrong code.
DemoThen, point your agent at the spec and watch it build to the blueprint. When it's done, use open spec to verify the work against the spec. And here's what's genuinely exciting.
With open spec, spec don't disappear after a feature ships. They compound over time. A living record of what your system does and why.
Every new feature builds on what came before. And when code drifts from the original intent, you'll know. All of this works locally, as Markdown files, checkable interversion control, right alongside your code base.
How It WorksNo API keys or MCP required. We work across over 20 different coding agents, anywhere agent skills work. And at 27,000 stars on GitHub, we're helping define how developers work with coding agents.
CloseLoved by everyone from solo builders to y-combinator startups to some of the largest engineering teams in the world. Check us out at openspec.dev.
Hook$760 billion are wasted each year to operational inefficiencies in healthcare. Most systems store lia on manual and dis-joinited workflows for staffing, patient flow, and corporations. We are scheduling wizard and we are building the logistics infrastructure to modernize healthcare operations, beginning with scheduling in order to combat this rampant inefficiencies.
SolutionIn the last year alone, departments at major hospitals including Mass General YouTube Southwestern, H-E-A-L-A-Generol and our alma mater Johns Hopkins have outsourced their physician scheduling to us and we are actively expanding. Our scheduling service is activated 18 departments across 13 hospitals.
How It WorksWe save our clients hundreds of hours by creating the schedules for you. How do we do that? We have a proprietary scheduling programming language and AI-driven workflows that allow us to create even the most complex schedules far faster than traditional methods.
TractionThrough this work, we have seen firsthand that scheduling is one piece of a much larger operational problem than healthcare and that's why we are building patient care tracking and provider management systems with design partners at UCSF and Harvard-affiliated departments. We aim to complement existing tools and EHRs to streamline clinical care coordination and add the logistical and predictive capabilities on top.
CTAOur goal is to unify infrastructure that incorporates the power of optimization, advanced statistics and mathematical modeling to reduce administrative burden and improve care coordination. Come join us in bringing healthcare operations into the future.
SolutionWe are scheduling wizard with the healthcare inefficiencies disappear.
Hookshortkit (YC W26) helps every company roll their own TikTok-quality video feed.
ProblemShort form video has become the default way that people experience the internet, but the value is captive on three big players. The best publishers and consumer apps (like Doordash and NYT) are launching video experiences in their own apps to capture engagement, first-party data, and ad spend.
How It WorksNow, with ShortKit, teams can roll out best-in-class feeds without massive video engineering teams. Their managed SDKs and video infrastructure are built specifically for vertical video and optimized for the buttery smooth experiences consumers have come to expect.
Michael spent years building video infra at YouTube, and now the team is bringing those same capabilities to publishers and consumer apps everywhere.
CTABook a demo at shortkit.dev.
Congrats on the launch, Michael Seleman and Neil B.!
https://lnkd.in/gF6QiJJe
HookShort form video has become the default way that people experience the internet. For years, three platforms had a monopoly. Now the best consumer apps and publishers are taking that engagement back.
ProblemBut it's still really hard to get right. Consumers expect funerary smooth performance, imperceivable latency, and wreck systems that know exactly what they want. To do all of that, you'll spend millions on infrecies.
TeamI spent years at YouTube building the infrastructure that powers YouTube shorts. Now we're making enterprise quality short form video systems easy for everyone else. Product teams can use our plug and play SDKs, video infrastructure, and wreck systems to deliver native short form video experiences that rival the big platforms.
Even better, we make it easy for you to build and deliver new experiences on top of short form, whether you're a housing marketplace, a language learning app, or a news media publisher. We're empowering the open internet with the same infrastructure that underpins the most engaging platforms in the world.
TractionWe're currently collaborating with leading teams across consumer and publishers. And if you're a product or growth leader who's thinking about video, we'd love to help.
HookSynthetic Sciences (YC W26) is building AI co-scientists for end-to-end scientific research.
ProblemScientific research is still a marathon of fragmented work across papers, code, compute, and writing. Each iteration forces researchers to rebuild context, rerun experiments, and stitch results back into drafts. You can't parallelize yourself, so progress bottlenecks on attention and context.
SolutionSynthetic Sciences lets scientists delegate the full research loop to AI co-scientists: literature review and synthesis grounded in your project, hypothesis generation, experiment planning, code execution in containerized environments, GPU runs, monitoring, analysis, and publication-ready drafts (LaTeX, figures, slides).
How It WorksThey've battle-tested synsci most on ML research so far and are rapidly expanding across domains. Biology mode is already SOTA on BixBench Verified (92%).
They're also launching Flywheel Mode, which helps teams turn production traces and data into custom, private, task-specific models they can own and improve over time. Thanks to strong open-weight models and advances in post-training and RL, these models often end up faster, smarter, and cheaper than frontier model APIs for their specific workflows.
CTATry it or book a demo: syntheticsciences.ai
Congrats on the launch, Ishaan Gangwani and Aayam Bansal!
https://lnkd.in/gU_u7kAq
HookHey, I'm Ashan and I'm Aya from some better sciences. Today, very excited to introduce you to the SunSide, our platform that lets you orchestrate AI-CoSignors. Let's see you to that ship.
DemoI'm going to ask it if we can find you an ESM due to predict protein mutations accurately not from here, it's in the driver's seat. After making a step-by-step plan, it hits the literature, reads through dozens of papers, and adjusts its approach based on what it finds.
Then, it designs experiment. Estimates compute costs and asks for approval. One ready, it spins up GPUs and launches the training job in model.
This can act safe so you can watch what happens in real time. Training finishes, corrupt 10 points on speed limit. That's solid and competitive with models 5-20 times larger.
It also writes the whole thing up for you. Tabels, methods, schematics and citations all verified and ready to go. Let's do math for a second.
ProblemToday, Strahopopin-Vate models allow you to exceed frontier performance and specific tasks through advances in both training and our own. But that's beyond us. Let's just have a look at this.
DemoWhat is it? It was frictionless. Watch, I just wish to fly with a model.
How It WorksThis is a harness for post training. Dawson's is computing integration, hundreds of skills. It orchestrates ML scientists to deploy that same model we just trained as a live API.
Any bad luck in pairing with every result feeding back into retraining. In practice, our AI co-scientists let smartines finally on the full loop. Train, evaluate, deploy, iterate.
Your model lives on your infrastructure. Get sharper with your data. And the advantage finally accrues to you.
This is an exciting problem. And a rapidly expanding across domains. For example, our biology model achieves state-of-the-art on big-svenge.
CTAIf you want to try our own specific theoretical scientists, visit synthetic sciences.au.
HookCarrot Labs (YC W26) builds custom models tuned to your tasks that never stop improving.
Every company building AI agents faces the same trap.
SolutionFrontier models deliver great quality but come with brutal latency. Smaller models are fast, but can't get the job done. You tweak system prompts until you hit a ceiling, and every time a new model drops, your prompts break. Carrot Labs fixes this.
Their platform captures your agent activity and tunes your custom model to your tasks from an open source base model. Swap out the frontier model for yours, and as new activity flows in, they continuously retrain your model to boost performance.
This is how you get a model built for your tasks that never stops improving.
CTACarrot Labs is already powering customers' agents! Get started today at carrotlabs.ai or reach out to contact@carrotlabs.ai.
Congrats on the launch, Christopher Acker and Yuta Baba!
https://lnkd.in/gkqz3Yp5
Every team is building with the same models. The same performance. Same prompt acts. And the same problems tomorrow. To solve this, carrot labs builds proprietary models that don't hit the same ceiling of the same darkness.
HookHi, I'm Ruth Fick and today we're launching Zetana. Zetana is a new way AI agents coordinate with operational software systems of record. For too long, AI agents have been gate kept from the web and software workflows needed to take real business actions.
ProblemThat's because that context allows us to have legacy software like SAP, point of sales systems, EHR and EMS platforms and insurance websites that businesses rely on every day to work. Today, companies that want their AI agents to interact with these software have to build and maintain their own custom automation and integration infrastructure themselves.
We build Zetana to solve this problem. Simply shows Zetana your workflow and our internal AI agents will understand the system, discover the underlying requests and build the series of calls required to complete the action programmatically.
How It WorksZetana then runs your workflow reliably in production. We handle IP rotation, TLS, credential and session management and anti-buff protections so that developers don't have to rely on brittle browser automation solutions that build themselves.
TractionWe already support customers doing millions of revenue, or connecting to external software is core to their product. Our systems handle millions of requests per hour that previously require the overhead of custom scraping infrastructure.
We all previously worked at high-growth companies where we had to go through the brutal process of creating our own scraping infrastructure. I previously built scraping solutions for several high-growth startups that handled hundreds of millions of requests per month and dealt with low latency and high-through button aids.
If you are working with legacy software or struggling with your scraping or integration needs, by the way, websites, desktop apps, or on-prem servers, we'd love to talk.
HookCarson AI (YC W26) deploys AI agents with enterprise-grade security to generate custom, task-specific interfaces.
ProblemAI tools today sit at two ends of a spectrum: software like OpenClaw is powerful but insecure, while alternatives like Cowork are useful but suffer from heavy context-switching and a clunky permissions system.
How It WorksCarson doesn't compromise. It's safe to deploy at work because the platform is built around a fine-grained authorization model. Yet it's also powerful, generating dynamic interfaces like research dashboards and slide editors customized to your current task. Built for problem solvers pushing the bounds of AI in sales, marketing, and ops.
CTASign up at usecarson.com.
Congrats on the launch, Sidharth Menon, Alex I., Ketan Agrawal, Milan Bhandari!
https://lnkd.in/gjVx9J6M
HookHi, I'm Sid and I'm very excited today to introduce you to Carson, a new generation of proactive AI agent with a novel, generative interface that adapts to you as you use it. Carson deploys teams of agents across your entire stack, stack of deliverables and run persistent workflows from Google Drive, notion, email and more.
SolutionLet me show you how it works. The home page of Carson is a familiar chat interface. Today, I'm a salesperson.
As can Carson to help me find some due leads. Carson starts running some deep research and transforms the screen in front of you into a spreadsheet since that's the best way to show the data. In order to fill in the table, Carson does web research and also utilizes best in class APIs for lead enrichment.
Once the research is complete, I ask Carson to help me with some outreach to these candidates. It transforms the screen into a specialized workspace for this build for just this task. On the left, hands out of the screen, there's an email client.
In the middle, the candidate's LinkedIn profile, run the right Carson's research. With Carson, you don't have to balance around between different applications anymore. I write the first email to Zany.
And once I'm happy with it, I set it off. Then, I ask Carson to turn that email into a template. I can now just press tab to hop between the relevant phrases in the email and write some personalized copy.
Carson is also smart enough to look beyond the surface level in its research on the right. It not only pulled Alberto's professional interests, but also looked for blog posts, social media presence and hobbies.
TractionIn this case, I see that Alberto and I share a hobby, so I add that in there too. The next day, I see Alberto as responded to my email and book some time on my calendar for Monday. I ask Carson to help by generating a slide deck for this first meeting.
Carson starts by deeply researching Alberto and his company. And once that's done, it starts generating slides. Carson is intentional about layout and branding.
How It WorksEvery slide has my company's branding as well as our potential clients. Furthermore, Carson gives you an easy way to edit your slides. Every element and asset on each slide is clickable, and you can either change the text directly or leave comments like you would for a coworker.
On this one, I have some feedback and I tell Carson to address it. Now that I'm happy with his slides, I ask Carson to automate the entire work slow. Every time someone responds and puts him in my calendar, generates slides just like these and send them to me.
Carson is built on a fine-grained permissions model, so it first confirms its authorization scopes with me to act autonomously. To visualize the automation, Carson turns the screen into a visual workflow builder where each node is plain English.
I have a small edit and asked Carson to ping me in Slack too, and then I deploy the workflow. The next day, I see Amit book sometime, and now we can see the workflow in action. Carson thinks me on Slack with a personalized presentation attached, just like I asked.
Carson is powerful, but it's also safe to use at work because of advances we've made in security for agentic systems. The entire platform is built around a fine-grained permission system, which means you can feel comfortable embedding Carson in your most mission-critical workflows.
TractionCarson is available today as a research preview. If you would like to try Carson in your business, please let us know. We look forward to working with you.
HookSonarly (YC W26) is your AI engineer for production.
How It WorksIt automatically triages production alerts, investigates root causes, and fixes bugs in the background.
ProblemMost teams don't trust their alerts anymore because of the noise. They end up relying on user feedback (too late) or manual digging (too slow).
This hurts users who have to wait hours for a fix. But it’s not because teams don't care, it’s because triaging and investigating every issue is manual, tedious, and requires hours of focus.
SolutionSonarly fixes that.
TractionYou can connect your monitoring stack (like Sentry or Datadog) and your user feedback channels (like in‑app bug reports or support tickets).
Sonarly automatically triages all these signals, links them to real user impact, investigates root causes, and ships a proposed fix.
If you're a startup or fast-growing company that cares about bug-free software and speed, try Sonarly, and cut your MTTR.
Congrats on the launch Dimittri Choudhury and Alexandre Klobb!
https://sonarly.com
https://lnkd.in/gdpXf-ci
HookHow do I sleep better at night? Why are you stressed? I'm an on-call engineer.
ContextMy phone never shuts up. Alert non-stop. Always something to fix.
I just can't keep up anymore. Did you hear about so gnarly? Hi, I'm Jimmy Trig.
SolutionI'm Alex. We are building sonali, the AI engineer for production. This year, most of where the ever is being created.
ProblemBig time has changed, but turn time is taking the best. When you are a non-colon engineer, you keep receiving noisy production alerts. You do time switching between all the different observability tools.
How It WorksYou spend hours debugging the issue while you are using our waiting for effects. We sonali, you can connect your monitoring stack. When an alert fires, sonali charges and prioritizes the issue.
It's automatically investigated for the wood course by looking traces, logs, metrics and your cut. We help you win hours of research and make sure to trust through our system. Behind the scenes, sonali updates and internal map of your predictions system.
In understands, I will share with you a better and more detailed analysis. I will date a flow of the pressure stack and how failures propagate learning from your observability data, cut changes and past incidents.
At any moment, you can jump in, ask forward questions and guide the AI agent. And because everything is different, sonali is fully customizable. You control the context it has access to and what are your team best practices.
DemoOur mission is to catch your own teacher so you can focus on building. We will be happy to show you the demo and our module in minutes. Try sonali.com.
HookSOX audits don’t have to consume your team -- or your budget.
Oxus (YC W26) is the AI-native platform that automates the most time-intensive parts of SOX audits.
ProblemToday, internal audit teams remain constrained by manual, repetitive tasks: sample-based control testing, walkthrough flowchart documentation, spreadsheet reconciliations, etc. This limits bandwidth for strategic judgment and insight. The alternative is costly outsourcing, with heavy back-and-forth and inconsistent quality.
How It WorksOxus automatically surfaces high-risk areas, generates process flowcharts, performs control testing, and produces clear, traceable documentation that aligns with auditor standards. With Oxus, internal audit teams go from raw evidence to review-ready results in a matter of minutes.
Congrats on the launch Janet L., Christine Watts, and Melinda Liu!
Hookashr (YC W26) automatically validates your agents with tests that reflect how users really use your product.
How It WorksAI agents are everywhere: legacy companies are scrambling to integrate AI into old software, and emerging AI-native companies are looking to transform entire verticals. By and large, these transformations fail, not because models aren’t good enough, but because agent errors, hallucinations, and lapses in quality have begun to chip away at broader trust. The reactive nature of AI testing is the biggest culprit to lost trust; developers wait for agents to fail in front of real customers, giving users the impression of a broken system. Ashr solves this.
SolutionTeams use Ashr for a variety of use-cases spanning testing voice agents to session-replay analytics software. Ashr has already saved teams the cost of human-annotated test sets, while giving them the tools to quickly iterate to minimize errors and ensure high-quality responses. Developers benefit from using Ashr to benchmark their agent’s reliability, providing much-needed assurance to clients. Ashr is already used today by some of the fastest-moving companies in current and past YC batches.
Get a free agent evaluation by booking a call at ashr.io
Congrats on the launch, Rohan Kulkarni and Shreyas Kaps!
https://lnkd.in/gnCbbGgk
HookYou building AI agents, you test them by chatting with them, compting them, manually clicking around, it seems like it works, so you ship. Then you'll really use their shows up. They use your product normally, then it breaks.
DemoYour team finds out in production. Users don't behave like your internal demo, and when an agent fails, it's rarely screw on the insert, in skipping a tool called calling a wrong function, taking actions completely out of order.
To users, that doesn't really like just a small bug. It looks like a part of they can't be trusted. They get frustrated, and the next tab they open is that your health dogs.
It's a competitor. On Trace, and I'm Rohok, and we're the founders of Aster. We call agent to systems in finance, DevOps, and voice.
ProblemIn recent agents pass every unit test of bug bash, and then fail in front of real users. We just got back by Y-commonitor, it fixed one of the biggest problems in building agents today. Test it.
We evaluate your agents, with large volumes of high coverage, realistic test scenarios, and then tell you where your system is work, and go. Asher tests AI agents programmatically. You tell Asher what your agent is supposed to do, the job, the system prompt, the tools, and functions it can call.
TractionThe inputs it takes, and the personas and use cases you care about. Plus any sample data you already have, then Asher generates the tests we wish you had time to write. Text scenarios, audio inputs, even simulated environments, built to hit real user flows, not just the happy path.
How It WorksThen you install the SDK in your code base. The SDK runs your agent against our custom data sets, and captures what really happened. What tools it called, what it skipped, and what it returned.
TractionSo you're testing real system behavior, not guessing from a handful of prompts. With Asher, you get a clear run summary of what passed, what fail, what could have been better. You fix issues before your user's payment, before tickets pile up, before trust splongers, and before retention ultimately takes the hit.
SolutionAs agents take on real responsibility, the experience is the product. And broken experiences don't get second chances. That's why this era of software uses new way to test.
TractionInstead of hoping that your manual testing gives you enough coverage, you generate custom tests, especially designed to probe new features, and catch bugs that you couldn't even predict. Now you sell birding hours, now you ship knowing what will fail, and you fix it before customers, I re-enotace.
If you're building an engine system, but you call it us, and get a free evaluation of your agents. We'll show you where your agents fail before your users do.
HookArcline (YC W26) provides AI-native legal services for startups, with same-day turnaround & elite lawyers in the loop.
ProblemArcline’s founder Pamir Ehsas, ex-outside counsel to OpenAI, saw the same issues on repeat when lawyers were serving startups: simple legal work took weeks, pricing was opaque, and lawyers kept starting from scratch instead of using AI.
SolutionThey’ve built a bench of top-tier startup lawyers to change that - folks from top schools (Harvard, Stanford, Oxford) and leading firms (Cooley, Goodwin, Fenwick).
Arcline’s lawyers start with first drafts generated with AI, and then do final revisions. This ends up reducing the work by up to 80% - so startups pay elite lawyers only for the final 20%.
More importantly, your legal matters can often be handled same-day. No other law firm today offers this kind of speed.
TractionThey’ve already scaled to working with 50+ venture-backed startups across the US and Nordics.
Congrats on the launch Pamir Ehsas and Stefan Mandaric!
arcline-ai.com
https://lnkd.in/ePFWk35a
HookOkay guys, let's do a rehearsal on the elevator pitch. Our client helps startups with the legal problems. Our combination of AI, Lawyer and the Loop provides legal faster and more affordable.
ContextSo they can focus on building wonderful companies. Start up scant afford having lawyers from day one. And we just need to change that.
We can change everything. 10 so thousand dollars were saved just because of the service. And in the future I think this could save us millions.
ProblemThat's crazy. Yeah, it is. People have been trying to use AI in a long time, but it hasn't been until now that we really had the technology to make a difference.
So we're going to take a unique approach. We're going to build a hybrid law company. So they can focus on building and growth.
So just like having a in-house legal counsel. Imagine whenever you need a shareholder agreement, founder agreement, master of service of agreement, you're coming to us and we're going to provide it to you within a few hours to every founder out there.
And with our hybrid approach, we can give them that. And making legal for them more affordable at the same time that is called the shirt by top lawyers. The worst comes from the biggest legal institutions in the world.
TractionThey work with founders that have been backed by the biggest vcs in the game. But now is really going to change the projection with black commentators as well. So we're moving to San Francisco to help a bunch of founders in the Bay Area.
Affordable quality from someone you can trust. So far I'm very impressed. I'm sure that they will become a big unicorn from Norway.
Finally. That's going to be very interesting, right? Yep.
ProblemA man of mayors. I think we are on to something with this hybrid approach. Because right now our legal system is broken.
HookKyten (YC W26) designs and manufactures custom aerospace-grade battery packs for defense applications.
Solution100,000+ drones, submarines, and satellites are being built in the US, but the legacy supply chain doesn't think like a modern manufacturer; they can't keep up.
TeamDuring their six years at Starlink, Cooper McBride & Lucas Maddox put 5,000+ battery packs into space. Now, they're bringing rapid development and volume production of aerospace-grade battery packs to the entire industry.
CTAKyten has capacity for a few pilot partnerships. Reach out now if your program needs custom aerospace-grade battery packs: founders@kytentech.com
Congrats on the launch, Cooper and Lucas!
https://lnkd.in/gdX_9YBu
HookAmerican Defense Manufacturers are building hundreds of thousands of autonomous vehicles for the air, sea, and space. But they all face an existential risk, a legacy supply chain that can't keep up, which is especially true for battery packs.
ProblemEvery modern defense system needs a custom battery pack, and one defect can destroy an entire production facility. America needs hundreds of thousands of safe aerospace-grade packs, and fast. Yet, our defense supply chain delivers low volumes of over-priced packs in years, not weeks.
And while America waits, our adversary's bill that scale. I've started like in Gild AI, Lucas and I spend a combined eight years scaling production of aerospace electrical systems that power over half the satellite and orbit, and hundreds of drones on the battlefield.
CTANow, we're building Titan, an aerospace-grade battery pack supplier built for speed and scale, so defense manufacturers can move fast without their energy systems holding them back. Visit cightintech.com to order battery packs now and see how we're reinventing the Tier 1 supply chain.
HookMaven (YC W26) lets voice AI agents collect payments during live calls.
ProblemVoice AI has been capable of handling real customer conversations for a while now. Agents can qualify leads, book appointments, and negotiate pricing. But when a customer says, “I’m ready to pay,” everything breaks.
How It WorksTaking payment over the phone means secure card handling, retry logic, gateway integrations, and PCI compliance requirements, all in the middle of a live call. Most teams either avoid payments entirely or take on the complexity of building and maintaining their own payment infrastructure.
TractionMaven handles payments for voice agents with a single API call. When a customer is ready to pay, the agent triggers Maven and it handles the payment flow end to end.
Congrats on the launch Wasi Ahmed and Brandon Boehme!
https://trymaven.com/
https://lnkd.in/g4hQH6tu
HookThis mission won't be easy. Here's the game plan. Voice Agent One, blend in with the people and get past security.
ContextSound human and that human. Step two, keep control and handle objections. Calmly.
Sorry, authorized personnel only. I am authorized. My apologies.
Please proceed. Step three. And listen closely, this is the important part.
Retrieve the payment impossible. We can't process payments. We need PCI compliance.
TractionThat's going to take three months! You can't. But I can.
How It WorksI'm dropping in a single API call. It's the universal key. It bridges the agent, FAPI, live kit, or Twilio.
Straight to the vault through Stripe, authorized.net, and more. Voice, dial pad, or SMS, we're secure. We're PCI compliant.
CloseAnd we get you paid. Mavin is how voice agents secure payments.
HookWhen your company raises a new round, you get a lot of money, like a lot of money. Sure, you need some of that right now, but most of that money is just going to sit there for months. That's an opportunity.
ProblemNow, Google, Apple, Amazon. These big companies have entire treasure teams dedicated to making every dollar work as hard as possible. But what if you're a startup?
What are your options? When we joined YC, like all our peers, we were bombarding by emails from breks, mercury, row, all of these companies expecting us to put money in their treasury platforms. But what do they actually do?
How It WorksThey just park your cash in a money market fund, collect their management feed, and forget you exist. And if we face it, that means you face it. That means every founder out there raising capital has already faced it or will face it in the near future.
SolutionAnd to solve this, today we are launching Palace Finance. Our solution is simple, higher yields, lower fees, and real investment strategies. All package seamlessly in a nap that hooks up straight to your existing bank account.
TractionPalace has portfolios run by professional investment advisors that optimize for what startups actually need, keeping your idle cash safe and accessible. All while earning up to 50% more yield than generic money market funds.
That can mean extra hundreds of thousands of dollars every year. Just for letting your money sit in the right place. At Palace, we aim to provide the kind of sophisticated treasury strategy that big companies use to every startup and business regardless of scale.
CloseSo that founders can focus on doing what they do best. Building.
HookConstellation Space (YC W26) predicts satellite network failures before they happen and automatically reroutes traffic to prevent outages. 🛰️
SolutionThere are 10,000 satellites in orbit today, and there will be 70,000 by 2030. Network failures already cost the industry $2.5 billion per year. The problem is that satellite operators are stuck in reactive mode—something breaks, engineers scramble, data gets lost, customers get upset. The tools they use were built for a world with dozens of satellites, not thousands. As constellations grow, no human team can monitor and respond fast enough.
How It WorksConstellation fixes this by analyzing telemetry from satellites, ground stations, and weather systems in real time. We use machine learning to predict link failures 5 minutes before they happen—at 90%+ accuracy—and then automatically reroute traffic in under 2 seconds. No human intervention, no data loss. Think of it as autopilot for satellite networks.
TeamOur team has worked on this exact problem at SpaceX, Blue Origin, and NASA. We've spent years building and operating satellite infrastructure, and we've seen firsthand where the current tools break down. We're building the software we wished we had.
Congrats on the launch Kamran Majid, Raaid K., Omeed Tehrani, Laith Altarabishi!
https://lnkd.in/euziWcFw
HookThere are 10,000 satellites in orbit today, and there will be over 70,000 by 2030. Network failures already cost the industry over 2.5 billion dollars a year, and the old model of managing these networks manually doesn't scale to thousands of satellites.
Consolation OS ingests so-editary from satellites, ground stations, and weather systems over 100,000 messages per second. We use advanced AI techniques to predict link failures before they happen at a greater than 90% accuracy rate, and then autonomously re-wrote traffic in under 2 seconds.
TractionZero data loss, zero human intervention. Our team has worked on this exact problem for years at SpaceX, Blue Origin, and NASA. We're testing with defense and commercial partners now.
CloseIf you're operating a satellite constellation, we'd love to talk. We're constellation, and we're building the operating system for space.
HookBubble Lab (YC W26) supercharges your ops work in Slack. Deploy Pearl in one click, connect it to tools like Notion, Jira, Stripe, and let it run tasks and automations for your team directly in Slack.
How It WorksOperations teams today spend endless hours switching between tools, updating systems, running reports, and coordinating workflows manually. Automating these processes typically requires engineering time or maintaining fragile integrations, which slows teams down.
Bubble Lab solves this by letting teams deploy Pearl directly into Slack in under a minute. Pearl connects to your tools and takes action on your behalf—updating HubSpot after a sales call, creating Jira tickets from user feedback, or generating daily revenue reports automatically.
TractionAlready powering 4,000+ operators and fast-growing teams, Bubble Lab helps teams scale their operations without scaling headcount.
CTABook a demo at https://bubblelab.ai
Congrats on the launch, Selina Li and Zach Zhong!
https://lnkd.in/duuvCVyx
HookI have never seen you leave Sluck. You're running our data reports chasing down unpaid invoices, scheduling follow-up meetings with our candidates. It takes me days of intense focus to do the same exact thing!
ContextLook, this is ridiculous. There's only one possible explanation. You are a witch!
Ha ha ha ha ha! What is so funny? Oh my god.
Look, you're like never heard a pearl. Oh, what is that? One of your witchcraft thingies?
How It WorksNo! It's your team's AI assistant living right in Slack. It has native integrations with basically every tool in your stack.
TractionHa ha, Jera, Nocean, Google Drive, Stripe. I can trigger automation, staff emails, update CRM reports, pull pipeline reports, all without ever leaving Slack. Pearl handles it all for you.
How It WorksSo, it's not witchcraft? No! It's like having an automation engineer on call-book up.
The reason it takes you days is because you're doing everything manually across 10 different tools. Pearl handles work in Processor Stack instantly. And when something's worth repeating, it turns into a robust reusable workflow.
Slack is the interface. Pearl is the brain. It takes 30 seconds to set up and get smarter over time.
CloseIt would set this up too instead of wasting everyone's time. No, I'm timing.
HookMouseCat (YC W26) is building AI agents for fraud investigations.
How It WorksMost fraud systems rely on static rules and ML models that are slow to adapt when patterns shift. MouseCat's agents investigate like a human analyst — reviewing case data, searching databases and APIs, referencing prior cases — but they do it continuously and at scale.
The result: faster detection of emerging fraud trends, fewer false positives, and agents that improve with every case.
People don't generally rave about infrastructure, but one of the coolest aspects of MouseCat is what's under the hood — built-in evaluations, monitoring, tuning, and compliance that help companies actually deploy and maintain their agents in production.
Nicholas Aldridge is a Core Maintainer of MCP and a former Principal Engineer at AWS. Joseph McAllister worked on ML infrastructure and fraud detection models at Coinbase and exited his first startup while attending Cornell.
Congrats to the team on the launch!
🚀 https://lnkd.in/gE-JH_G9
HookFROAD isn't new. AI enabled fraud is. Mousecat uses AI agents to perform human quality fraud investigation at scale.
How It WorksWe work behind the scenes following your standard process, analyzing each case and surfacing the most important evidence. We scale human grade review to discover new fraud patterns 24-7 and stop them immediately.
HookVerdex (YC W26) is replacing physical inspections in insurance with real-time satellite imagery audits, allowing insurers to verify and settle claims without ever leaving their desks.
ProblemCrop insurers spend over $5 billion annually on a single manual task: claim verification. Currently, insurers are dependent on physically sending adjusters to fields to verify losses, a process so inefficient that it consumes nearly 10% of every premium dollar collected globally. As climate volatility continues leading to a significant rise in claim volumes, the current adjuster model is struggling to keep pace with the increasing demand.
TractionVerdex solves this by enabling insurers to monitor fields in real time through a fusion of high-resolution satellite imagery and local weather models. By interpreting field-level conditions with high precision, their platform allows insurers to verify assets in seconds and settle claims instantly, replacing weeks of waiting for a physical visit. Recent regulatory shifts have now made this transition possible, and Verdex is already live with one of the largest crop insurers in the US, helping insure millions of acres of American farmland.
Congrats on the launch, Jad Bousselham and Evan Rankin!
https://lnkd.in/g67-Rr_e
HookHi everyone, I'm Adam. And I'm Jed. We're building Vertex.
ProblemEvery year, crop insurance worldwide spend over $5 billion on a signal poll. Manual verification. Today, in Chorus Camera Verify claimed with a physically sending an adjuster to the field.
SolutionThis process is so inefficient that it eats up nearly 10% of every premium dollar worldwide. Vertex replaces physical inspections with a real-time digital audit. We've used hybrid solution satellite imagery with local weather models to monitor every field on Earth, 24th Saturn, allowing insurance to settle the same without ever-leading a desk.
TractionOur timing couldn't be better. Last month, major regulators, including the US government have updated their guidelines to allow for satellite-based remote verification. We've already launched with one of the largest crop insurance in the US.
I'll be even sure over 100 million acres. That's roughly 11% of all US farmland. This is just a start.
We're digitizing a $50 billion dollar market that is bubbling in size to claim it volatility. If you want to see how we're automating the future of insurance verification, let's talk.
HookSo, rapid deployment, self-sufficient data centers. Good summary. Tell me more.
ContextAI is consuming an incredible amount of energy. Utilities can't keep up. Multi-billion dollar projects.
ProblemI have to wait over five years just for a power connection. Data centers are being forced to solve this problem themselves. Our philosophy, the best grid connection, is no grid connection.
Solar embattories on site, no waiting. And you can produce that much power. Oh yeah.
TeamWe can build into the hundreds of megawatts. Solar and battery storage have reached the threshold where there are no longer just an expensive alternative. They are becoming the de facto solution.
How It WorksFor the better part of the decade, we've been developing the supply chains and the technology, so we can deploy this at scale map. We're developing a standard. We've gotten ahead of the curve by using second-life EV batteries to develop a proprietary power distribution system that is not only faster to deploy, but more efficient and more redundant than existing technology.
How long does it take to actually deploy one of your data centers? How long have we been talking?
HookVisibl Semiconductors (YC W26) is building the first coordination layer for chip design.
As chip teams scale, critical workstreams drift out of sync and issues surface late, driving costly delays ($XM+) and months of rework before tapeout.
Visibl’s long-term vision is to build end-to-end AI systems that shrink chip design from years to weeks. The team is starting with coordination because that’s where design intent breaks first.
How It WorksVisibl was founded by Bryce Neil and Jordon Kashanchi, who have built production systems and custom AI chips at Microsoft, Intel Corporation, Arm, Deloitte OmniaAI, and SST. They started Visibl after repeatedly seeing coordination bottlenecks delay integration and tapeout.
SolutionSeeing Visibl in action will change how your team approaches chip design.
CTABook a demo at www.visiblsemi.com
Congrats on the launch Bryce Neil and Jordon Kashanchi!
https://lnkd.in/ghHgeqKZ
HookEvery chip design I've been through starts the same way. The people are experienced, thoughtful, and deeply invested in getting it right. But as designs grow, and more people get involved, keeping everyone aligned becomes a different kind of challenge.
ContextI've built chips across multiple organizations, and what becomes clear is teams tend to silo themselves as the complexity of the device grows. The opportunity is creating a shared picture that stays current as the design changes.
ProblemWe're living in a world with incredibly intelligent AI systems, and it's a generational moment for semiconductor design. Because for the first time, raw engineering output will no longer be the bottleneck.
We're betting that design coordination is the unsolved constraint on tape out cadence, and it's why we're building visible. The first software defined coordination layer for chip design. Visible combines requirements, architecture, and implementation into one living platform.
TractionAgents are instantly dispatched on nightly regressions, failing PR checks, and spec updates, while proactive agents scan for drift on a regular schedule. Engineers open the case, and visible has already investigated, read the spec, check the implementation, and identify the root cause.
Three fixed options, one recommended. Engineers review, select, and submit. The fix executes, files update, verification runs.
Unable then implement the fix, updates the relevant files, and triggers the pipeline to verify. Meanwhile, the executive sees what matters most. Schedule, readiness, and top tape out risks.
How It WorksWhen the engineer closes the case, the dashboard updates. Schedule impact resolved, both see the same truth at the level they need. Seeing visible in action will change how your team thinks about chip design.
HookTepali (YC W26) is building the all-in-one AI-native OS for medspas, replacing the disconnected tools medspas use today.
How It WorksMedspas are one of the fastest-growing industries in the U.S., with over 30,000 locations growing 20%+ year-over-year. Owners spend thousands on marketing but can't connect ad spend to patient lifetime value without pulling data from multiple platforms into spreadsheets. Current solutions weren't built for the AI era. Tepali builds AI into the foundation, with an intelligent assistant that answers any question about your business in seconds.
TractionThe founders have been best friends since middle school. Vishnu built all-in-one operational software for coffee shops at Dripos (YC W20). Chrisvin executed over $30B in M&A at PJT Partners and saw AI transform high-ticket service industries at Rilla. They're combining their deep technical and financial expertise to modernize how medspas operate.
Congrats on the launch Vishnu Pathmanaban and Chrisvin Jabamani!
https://lnkd.in/egK333d5
HookMedspos are one of the fastest-uerrex segments in healthcare, but the software is still playing capture. Medspos sit between a doctor's office and a day's spa. They provide treatments like Botox, laser therapy, and peptides.
SolutionBut current, all-in-one solutions weren't built for AI. So owners have to duct tape a dozen different pieces of software just to run their business. That's why we built a poly.
The all-in-one, anative operating system for Medspos. To poly connects your ad-spent to your customer lifetime value. Just ask the AI what campaigns to scale, and it'll tell you what's printing money, and what's to cut.
How It WorksEvery patient gets a single profile from first ad-click to latest treatment. The AI flags who's overdue for their next filler appointment, who's at the risk of turning, and what to upsell next. Your stock levels update automatically as you charge treatments.
The AI monitors your inventory levels, predicts when you'll run low, and re-orders before you even have to think about it. Every treatment gets us so in charge. Injection maps, before and after photos, consent forms, all-hip-up comply.
TractionThe AI autoattaches everything to the patient profile, so nothing gets lost between business. Because all your data lives in one place, you can just ask, what does my data look like? Why did revenue drop last month?
CTAAny question you have to polys got you covered. Your Medspos is their software built for 2026, not 2006. Visit to poly.com to learn more.
HookGeneral Legal (YC26) is an AI-native law firm focused on commercial contracts for growth-stage companies.
They provide high-quality contract review, negotiation, and drafting with fast turnaround and flat, all-inclusive pricing per contract. Their founders were AI/tech leaders at Casetext and practitioners at Cooley and Fenwick.
How It WorksGeneral Legal is rebuilding the law firm for the AI era like Tesla rebuilt the car for the EV era. They hire top-tier attorneys and equip them with proprietary AI workflows that enable them to operate at 10× the efficiency of traditional attorneys.
This tech advantage allows them to offer clients:
Traction- Flat, all-inclusive pricing: $500 per contract for review services and negotiation on existing paper
- Turnarounds in hours not days or weeks: Guaranteed contract turns within 3 hours, with typical turns delivered in 1 hour
How It Works- Highest-quality legal work: AI workflows eliminate common contracting errors and expert attorneys spend time focused on substantive risk and providing tailored advice
TractionSince announcing their private preview last month, they’ve had over 20 growth stage companies move commercial contracting to their firm. They're now publicly launching their commercial contracting legal services to serve any fast growing business.
Congrats on the launch Ryan Walker, Javed Qadrud-Din, and J.P. Mohler!
https://lnkd.in/gR4PVptc
HookTraditional law firms are slow, inefficient, expensive, and provide a guarantee of their own needs. I'm Ryan, I'm jumping, and I'm JP, and we're building generally an AI native law firm that delivers fast, high quality legal work at low and predictable prices.
SolutionWe've built AI products that are in the legal space at case tax where I was going to seek to get up. And JP and I graduate from Harvard Law School and we're at top firms, then like it truly is a choice.
TractionGeneral Legal is a full stack AI law firm, solving commercial contracting for founders. We say founders hundreds of dollars per contract. Our flat-ray fee, a $500, covers everything from red lines to face a face negotiation.
No billable hours, no retainers. We turn contracts in hours, not days or weeks, and our attorneys provide practical real-time guidance to get your deals close fast, but safe. You'll work in our experienced US bar to increase on either slack with your health.
And our AI will accelerate the work. We're already live in a completely different situation. So if you need help with your commercial contracts, you need to be done fast and correctly email us at foundersin general.com.
HookPollinate (YC W26) automates supply chain operations using your existing ERP. Custom supply chain tools, built specifically for your workflows, deployed in days instead of quarters.
Off-the-shelf supply chain tools force you to change how you work. Custom development takes forever. Your team gets stuck with neither, which leads to valuable operations running on spreadsheets.
SolutionPollinate reads your data structure, identifies your operational patterns, and builds the tools you need, to save you time across invoice processing, just in time production and vendor management.
Congrats on the launch Adeep Mitra and Corey Berther!
https://lnkd.in/gxHqgmUw
HookYour ERP.it isn't usable for your real world operations, and your supply chain teams are running on tools that won't design for them. And because billion dollars supply chains run on spreadsheets and emails, agents will struggle to leverage your fragmented data sets to drive real world visible value.
SolutionThat's why we created Polan, a platform where you can build and deploy AI agents that actually fit with your tools. Here's how it works. We plug into your existing tools and get your data ready.
TractionWe can spin up agents, specific fuel supply chain needs, without waiting 18 months for implementation. Here's a common example, where we pull invoices straight out of our customers' email inboxes. And then we pull them into Polan Head, where we automatically reconcile them, begins your patch disorders, and landing receipts from your ERP.
ProblemSaving on hundreds of hours of manual processing time. This invoice is now ready to go straight into your account existing. We also do order processing procurement management and demand planning to solve all of your most painful supply chain problems.
CTAGet more out of your ERP and reach out for a demo at Polan Head.
HookBurt (YC W26) helps teams train and deploy specialized models that outperform frontier, generalist models at a fraction of the cost and latency.
How It WorksAs good as general models are, they’re just not built for your use case. The fastest models aren’t fast enough and the best models don’t truly understand your domain. You’re paying unsustainable API costs while still being subject to unexpected regressions, service disruptions and annoying rate limits.
SolutionThe solution to that is to build task-specific models that are trained to do your exact task. Models you own, trained on your data, optimized for your use case.
How It WorksBurt has built these models for some of the fastest-growing AI companies using state-of-the-art post-training and inference techniques. They manage every step of the process from data prep and evals to training and deployment.
Congrats on the launch Bobby Zhong and Kurt Sharma!
https://lnkd.in/gXf9gFpJ
HookHey I'm Bobby, I'm Kurt and together we're a bird. Every team building a agents eventually runs into three problems with the models they use. They're either too slow, too expensive or just not good enough.
SolutionAs good as general close source models are, they just aren't built for your use case. The Excellency Generic benchmarks like Sweetbench or GSMAK, but your business is in a set of different benchmarks. They're decent at pretty much everything, but people don't actually care about everything.
A production scale in usage, teams are constantly choosing between using a big model that's good but slow or a small model that's fast but bad. Your APA costs are sustainable, your agents slow, and you have no control over a core piece of your product.
How It WorksThat's where we comment, we help you train and deploy custom models built specifically for your use case. We work with teams to build LLMs and VLMs that help perform general models while being 10 times faster and cheaper.
Models train on your data, optimize for your use case, and solving your problems. We help with every step of the process from data prep and emails to training and deploy. So if you have agents where LLMs caused that are too slow, too expensive, we're just not good enough.
CloseGive us a call at 1-800-tray-Memail. We're just having emails.
HookOverdrive Health (YC W26) is automating the entire medical billing and RCM lifecycle, starting with the $2bn ambulance billing industry.
ProblemHealthcare providers spend 2 to 10% of their revenue just to get paid for services they've already provided. The process is so manual and complex that they're stuck with two bad options: hire an in-house army of billers in an area they have no expertise in, or pay a third party up to 10% of revenue and hope they actually chase every dollar. Spoiler: they don't.
How It WorksOverdrive Health is an AI-native billing company that can actually afford to go after every single dollar a provider is owed. By automating insurance phone calls, documentation checks, benefit verification, and more, even a $100 claim becomes profitable to collect, meaning their incentives are truly aligned with their clients'. They pair their AI systems with a small team of expert billers to collect faster and more consistently than traditional billing companies.
TractionFounder Michael Schroeder has spent his career automating workflows in complex, operationally intensive businesses, first as an engineering lead at vertically integrated property manager Up&Up, then helping scale EliseAI from $35m to >$100m ARR. Overdrive is already serving >20 ambulance providers and collecting over $50mm in claims/yr.
Congrats on the launch, Michael!
https://lnkd.in/gvK5t7-G
HookThis ride is costing the ambulance company $2,500. They don't have a square wheel. They don't even know if I have insurance.
ContextSo how do they get paid? It's a 1990s nightmare. U.S.
Tractionmedical billing is a $200 billion industry. Yet providers lose up to 10% of revenue to armies of staff stuck on hold and sending taxes. That's literally setting money on fire.
Our AI agents are already processing thousands of ambulance rides every week. In just 90 days, we doubled the productivity of the traditional billing workforce. We automate the relentless busy work.
ProblemFreeing our expert human team to fight for the difficult claims that traditional agencies literally throw in the trash. We collect more money in less time at a fraction of the cost. We are over drive health.
HookDocura Health (YC W26) automates medical record review and med-legal report writing for workers’ compensation cases using AI.
SolutionIn workers’ comp, cases move at the speed of record review. Evaluators, claim adjusters, insurers, and attorneys manually sift through thousands of pages of medical records to extract timelines, injuries, prior history, and inconsistencies before a settlement can happen. Settlement often takes months, and the review work alone takes hours to days per case, creating the bottleneck that delays resolution and increases costs across the system.
TractionDocura transforms thousands of pages of records into structured chronologies and draft-ready, compliant reports in minutes instead of hours. It surfaces the key medical facts, organize them into clean timelines, and generate reports experts can quickly review and finalize. By automating the most time-consuming part of the workflow, it helps experts increase throughput and help insurers and claims teams reduce settlement timelines from months to days once review is complete.
Congrats on the launch, Akhil Sachdev!
https://lnkd.in/gJUKzf4j
HookAcross workers' complex, decisions are only a spouse's record review. And record review is still drowning in paperwork, boxes, of medical records, thousands of pages, endless PDFs, even CDs. In 2026, per position, this means spending hours reviewing records and writing reports.
ProblemFor attorneys, this means delays that slow decisions and settlements. For the injured worker, it often means waiting far longer than they should. Not because the work isn't being done well, but because the current system doesn't support it.
TractionDucura was built to change that. We transform thousands of pages of records into clear chronologies, structured record reviews, and compliant reports with built-in quality checks that reduce the risk of denials.
What used to take hours can now take minutes, and with delayed cases for months, can I move forward and weeks? Because understanding the case isn't the real breakthrough. It's delivering speed and accuracy from record intake to settlement.
CTAIf you're dealing with lots of records, Ducura can help. Visit ducurahelp.com.
HookPolymath (YC W26) builds simulated worlds for training & evaluating autonomous AI agents. We're heading towards a future where agents will be able to perform useful work over long horizons, with little or no human supervision. To increase the performance of autonomous agents, they must be trained in simulation environments that reflect the real world.
The team recently launched Horizon-SWE, a benchmark that drops frontier models into a simulated software company. The environment consists of a running application and real tools. Agents are given long-horizon tasks covering the entire software development lifecycle. Leading models scored ~25%. This benchmark measures the ability of AI agents to perform end-to-end agentic SWE tasks, as opposed to code generation alone.
TeamPolymath is a team of researchers and engineers from UC Berkeley, Hume AI, Plaid, and Amazon. The founding team has years of experience post-training frontier models in industry, and building large-scale data systems.
Congrats on the launch, Dylan Ma and Naren Yenuganti!
https://lnkd.in/grEZ-dGG
HookHi, I'm Dylan. I'm Nurey. A polymath, we're building the training grounds for AI agents.
ProblemAI models aren't just answering questions anymore. They're being asked to do real work and to end. But give them something that takes hours or days and they fall apart.
Static datasets are no longer sufficient. To train agents to do real work, you need environments. Worlds where Asians can practice, fail, and learn.
Traditionally, environment generation has been bottlenecked by human data. We've reached through contractors at the problem, hand-building artifacts one by one. We believe that human data alone is not enough to achieve super intelligence.
How It WorksWe're creating the core technology to enable automated environment generation using much less human effort than traditionally required. This allows us to build more complex and realistic worlds, and to achieve better quality, scale, and diversity of tasks.
TractionWe recently launched Horizon Suite, a benchmark that drops frontier models into a simulated software company. It consists of a running application, real tools, and long horizon tasks covering the entire software development lifecycle.
Frontier models were only able to complete around one in four tasks and to end. We're only scratching the surface of what's possible without autonomous agents. We believe that the future will have long-running agents in every discipline, performing at superhuman levels.
Founder at Anto (YC F25) | Researcher at Broad, Harvard, MIT | Lecturer at ETH Zurich
HookIn 1903, the first flight lasted 12 seconds. It took 60 years to go from Kitty Hawk to the sea of tranquility. By 1998, we started building a lab in low earth orbit and we've been there ever since, but astronaut time is limited.
ContextThe science waiting to be done isn't. Since the dawn of man, man's work has been limited from dawn to dusk. The future isn't sending more crew.
It's sending systems that don't need to sleep. Don't need to rotate home and don't have to pick and choose between which experiments to run. Microgravity changes what's possible.
Proteins crystallize with fewer defects. Sells grow without settling. Fibers form without the flaws that gravity creates.
Better pharmaceuticals. Pure semiconductors. Research that can't be done anywhere else.
SolutionThe bottleneck was never the science. It was who's there to do it. General astronautics.
HookMochaCare (YC W26) unlocks home care agency growth with human virtual assistants supercharged by AI, working faster and smarter on your behalf.
TeamThey take hiring, scheduling, or both off agencies' plates so they spend less time drowning in ops and more time building relationships in their communities. Co-founders Nick and Pranav personally work with each agency.
How It WorksToday, home care agencies are stretched thin: recruiting pipelines stall, caregivers call out at 2 AM, and growth opportunities slip through the cracks. It's hard to grow when you're constantly firefighting. Agencies pick one or both of their services: (1) Hiring that moves candidates from application to onboarding with AI interviewers, automated reminders/support, and human outreach for top candidates. (2) Scheduling that integrates into EHRs to handle call-outs, find coverage, and manage reminders and documentation, with a human in the loop for escalations.
TractionThe team is available up to 24/7 — you choose the hours. Everything they do generates data insights that surface growth opportunities, from referral leads to sales enablement during intake calls. For those not looking for a full service, their AI tools are also available as standalone software.
Nick Walker (Stanford BS/MS CS, ex-Spotify/Microsoft/Amazon/Linktree) and Pranav Uppili (UIUC MS CS, ex-GoDaddy) are both former caregivers raised by their grandparents. They started MochaCare because better ops means better care. On a personal note, Nick's grandmother started working with a home care agency and her caregiver didn't show up on day one. Not because the agency didn't care, but because running an agency is incredibly hard. They want to help so no family has to experience a loved one going without care, including Nick's grandma.
Congrats on the launch, Nick and Pranav!
https://lnkd.in/gcdf5DsA
HookWe met Shelley 20 minutes ago and her caregiver hasn't shown up yet. 10,000 Americans turn 65 every day and we don't have nearly enough caregivers. Come on people.
CTAIt's hard for care recipients, care agencies and caregivers who are all spread then. Behind every misvisit, there's an agency. Trying to keep up.
Hello, I'm Josie. You're your shift in 30 minutes. Can you make that?
I'll let you see what I can do. Hi, Keith. I wonder if you can pick up a shift now.
Yeah, really Sarah. Hi James. These are how a really tight slide.
Could you just get it? No, no. Yeah, I understand.
Yeah. Becky, well I need someone pretty away. Yeah, it's 11 am.
ProblemBut fine, I can go ahead and I'll talk to you. Okay. Thanks.
How do we fix this? Hi, I'm Bronov. We're Mowger Care.
We help home care agencies grow by taking scheduling, hiring or both off your plate. For hiring, we screened applicants and closed your best caregivers. For scheduling, we fill callouts and handle the day today.
Any time of day. Every agency gets a real person. Supercharge by our AI phone assistance.
AI interviewers. And more. And everything we do generates insights.
So you know exactly where to grow next. And together, we'll go from what you just saw to this. To this.
No worries about that. You're totally gonna talk to me. Well, I just wanted to comment to know what it could be.
How did you like that, Zima? You're eating an orange juice? I'm on your time.
I'm the way. No problem. I'm not a great son.
You're on to start over again. You get the baby. But you won't tell you.
TractionI'm in. I'm in for you, baby. Oh, you're there already.
Thank you. I'm in. Mowger Care.
Let's grow your agency. But wait. Before you go, we have to show you one last thing.
HookAt origin, we're building AI to make cell and gene copies safer for diseases like cancer, neurodegenerative disorders and more. They've been applied successfully to treat various disease indications, but their widespread application has been limited due to safety issues.
ProblemFor example, in cancer, Garty has been highly effective against multiple B cell malignancies, but its use in solid tumors is limited by off-to-short activity. Our model access is capable of designing malignative synthetic DNA switches and dials that program precise gene expression patterns ensuring that therapeutic effect is no-polised only to target cells.
How It WorksWe're still in the early days of cell and gene therapies. Using our AI design DNA sequences, we can encode logic that ensures hiding your subsidy and the ability to adapt the dynamic changes in the cellular environment that are allowing us to go after diseases that are more complex and harder to treat.
HookPerfectly (YC W26) is an AI-native recruiting agency that fills roles in days by automating sourcing, outreach, and screening.
How It WorksWhile building large ML orgs at TikTok and Meta, the founding team sat through over 800 agency-sourced interviews. Entire roadmaps would stall because roles stay open for months. After working as ML Scientists at TikTok on the world’s most advanced AI recommendation pipelines, they couldn’t believe how outdated the recruiting process was in comparison.
Perfectly uses AI to fill roles by automating the grunt work of identifying, reaching out to, and screening qualified candidates. Their ultimate obsession is one thing – providing the highest quality candidates.
TractionOne customer even fired all their other agencies two weeks into using Perfectly.
Their exclusive launch offer: they’ll undercut your cheapest agency by 50% (or give an exclusive discount) for your first role. Book a demo with them now and mention “Launch”: https://lnkd.in/e5-qyPmc
Congrats to Victor Luo and Zhuang (Gary) Luo on the launch!
https://lnkd.in/er2Q9iYD
HookWe're perfectly the first ever end to end AI recruiting agency. We fill your role in days, not months. You have big ambitions for your company, so you open head count and higher recruiting agency.
ProblemBut it's already been three months and your roles stay unfilled. That's because traditionally, Returning agencies are bottlenecks by a recruiters' time, so they're forced to trade off between speed and quality.
How It WorksPerfectly removes that bottleneck, so that candidate quality becomes our number one priority. Getting started is as simple as uploading your job and doing an intake through our portal. We're propped deeply into what you care about and your vision for the role.
TractionThen we'll run our algorithm on hundreds of thousands of candidates to surface those that are a perfect fit. That's something that was typically take a whole team weeks to do. Interested candidates will pass through a QIFI screen and once you're ready to interview, we'll send them straight to your slack.
Typically, we'll fill your interviewing schedule within the same day as you're in take. So far, we're filling four times faster and find candidates two times more precisely than traditional agencies for some of the Bay Area's fastest growing startups.
CTAWe're a team of ML scientists from TikTok, Instagram, and Facebook, and we're applying our recommendations to systems expertise, alongside some of recruiting's most prominent leaders. So if unfold rules are slowing down your team and you want to get back to building products, request access, we'd love to stick with you.
HookI am Matthew, the founder of drone tector, and we make radar that attacks small drones. Here we have the system in action. The difficulty around looking for small drones isn't the skies are actually quite busy, full of birds, insects and other flying objects.
ProblemExisting technologies struggle here, confusing some of these targets for drones. Drone tector can handle all of this, tracking up to 500 targets simultaneously. But where our sensors really excel is an identifying which target is a drone, which we do with near perfect accuracy.
Here a small drone flies towards one of our sensors at our testing site. You can just about see our system on the drone's camera feed is the white box. Our sensor detects all the targets around us, and after a few moments, tags that this target is actually a drone.
SolutionFrom this point onwards, the drone is continuously tracked. Current drone detection solutions make mistakes, commonly confusing birds and drones, and small drones are able to slip through undetected. Our technology solves this, providing much more reliable drone detection, the works against the most challenging targets.
DemoOur system is ready for live demonstrations and pre-orders, so if you experienced the drone threat, would love to hear from you.
HookHey, I'm Georgia. I'm an extreme athlete, turned software engineer and a repeat biomedical founder building mantis biotech. Here's the problem.
SolutionOver 80% of clinical trials are delayed and one of the biggest reasons is data quality. Each delay trial costs on average $15 million. So we built mantis to prevent these delays.
In 2024, an Alzheimer's drug called PTI 125 was discontinued during late-stage trials after questions about data traceability surface. By the time the program was stopped, hundreds of millions of dollars had been lost.
Years of time to market were gone and 7.2 million Americans were still waiting for treatment. This wasn't even a failure of science. This was a failure of data.
In most farm accompanied trial data is spread across EDCs, lab systems, analytics code, and vendor platforms. So when someone asks across system question, the answer isn't just a query. It's a full-on data project.
TractionIt takes weeks and involves multiple teams and is often cost hundreds of thousands of dollars just so the work gets thrown away afterward. Mantis is a domain aware data platform for life scientists that encodes biological, clinical, and human performance meaning directly into reusable data sets.
How It WorksIn the PTI 125 case, instead of weeks of coordination, a scientist could have just asked, show me patients with significant biomarker change, group by site and assay batch alongside cognitive outcomes, and gone and answered back immediately with full traceability back to the source data. We integrate across EDC, CTMS, labs, and omics platforms and produce canonical version data sets with full-enaged source systems.
CTANow, teams can answer questions, queries, and not projects. So, if you're a drug developer looking to reduce delays during FDA trials, shoot us an email.
HookJinba (YC W26) is a workflow builder for enterprise AI automation.
How It WorksThey serve 40,000+ users across Fortune 500 Global companies in highly regulated industries like banking and insurance. Enterprise teams want AI automation but keep hitting the same wall: they spend $300K+ outsourcing workflow automation to consultants and external engineers, only for projects to die before production because of security, scalability, and deployment requirements.
With Jinba, you describe the tool you need and your whole company has it.
Build AI workflows through chat with Jinba Flow, no engineers required. Deploy them as MCPs with one click so your team can start using them immediately through Jinba App. Role management, audit logging, SOC 2, on-prem: all handled.
Build something your colleagues want.
Congrats on the launch, Shoya Matsumori and Takuya Norisugi!
https://lnkd.in/gYg9tNFF
HookUsed by 40,000 enterprise employees from global Fortune 500 companies, Jinba allows anyone to build AI automations and easily share them across the organization. Build automations with chat. Edit with drag and drop.
How It WorksAnd perfect with code. Add as many guardrails as you want, for even the most challenging tasks. Create agents, autonomous workflows, and integrate with your existing systems.
With audit logging, on-prem installations, and access control, Jinba is built for enterprise. Build something your colleagues want with Jinba.
HookAemon (YC W26) is an AI R&D engineer that finds your team’s next breakthrough, automatically. Give it a problem and a way to measure success — it reads state-of-the-art research, then generates and evolves thousands of approaches at machine speed to discover optimal solutions, often going beyond what human experts can find.
SolutionGlobal R&D spend is $3.1 trillion a year. And every R&D problem has the same shape: you know what success looks like, but you don't know how to get there. Today, that means hiring R&D engineers to read papers, try approaches, and iterate — by hand. It's slow, expensive, and no team can explore more than a tiny fraction of the solution space.
Aemon does what a world-class R&D engineer does, but at machine speed and scale. Experts stay in control — a simple direction steers the search, making it faster and more precise.
TractionThe team pointed Aemon at circle packing — a classic NP-hard optimization challenge mathematicians have worked on for decades. Aemon set a new record, beating Google DeepMind's AlphaEvolve record, with less than $10 of compute. https://lnkd.in/dGk5kmy3
Give your R&D team superhuman reach. Book a discovery call at www.aemon.ai
Congrats on the launch, Yifei (Ray) Xu and Richard Zhou
https://lnkd.in/dtXThU8c
HookHey, I'm Ray from A-Monti and today I'm incredibly excited to share an early preview of A-Mont. The first AI R&D engineer. Let me show you what it looks like, an action.
ProblemI'm giving A-Mont a real problem here. A legal retrieval system is underperforming. Some test queries are failing on multi-hop cases.
I touch the failing cases and hit enter. First, A-Monti builds an evaluation harness. It analyzes test cases, identifies educators, categories, and runs a baseline.
We call a 10, setting at 54%. Before experimenting, A-Monti asks me to review the EVA. I check it, approve it, now we are aligned on what success looks like.
TractionA-Monti reads through the co-based on its own. It traces the retrieval pipeline, maps that are architecture, and pinpoint the root cause. The system does a single retrieval pass with no cross-dark in the reason.
Then, it does something no coding agent does. It reads recent researchers' papers, compares their approaches, and reasons about which strategy actually fits or co-based. And then it iterates.
Each attempt builds on the last. When something fails, A-Monti analyzes why in adapts. 23 iterations in, we call a 10 goes from 54% to 91%.
The result, production ready changes across three files. Not a prototype. A PRU can merge directly.
ProblemTo test out intelligent AMOIS, we pointed out that the circle packing problem, a decade-old MPHR that's medical or pre-mosition challenge. AMOIS and a new record being under 15 iterations. When they access some learns to hypothesize, experiment, and evolve the pace of discovery changes.
CloseFrom retrieval system to ranking algorithms on scientific simulations to financial models, we can't wait to see how AMOIS shapes R&D across every industry.
HookLuel (YC W26) is a rights-cleared data marketplace and collection engine for AI teams that need to-spec multimodal data fast.
How It WorksFrontier labs are hitting a wall: public web data is largely exhausted, synthetic-only pipelines can drift, and most “available” datasets don’t meet production standards for rights, provenance, and metadata.
SolutionLuel delivers custom collections with clean provenance. Teams submit a dataset spec (modality, scenario, instructions, devices, QA rules), and Luel scopes the project, recruits from a vetted global contributor network, runs multi-stage QA, and delivers licensed, audit-ready datasets within days.
Previous collections have ranged from patient-doctor conversations in South Asia to gemstone footage for robotics.
Congrats on the launch Inigo Lenderking and William Namgyal!
https://lnkd.in/gA9k_qwB
HookHi everyone, I'm William and I'm Inigo. We're building the well. From your labs have hit a wall.
ContextThe next generation of models, ones that see, hear, and act, require multi-modal data that simply does not exist as scale. The well will reliably turn the world's most valuable unindex asset in the world.
ProblemPeople's reality into multi-modal training data delivered within days. We mobilize our enormous global contributor network to capture their daily lives. Be it audio, persperson videos, images, or dashcam footage.
TractionAnd we're getting thousands of signups every single day from users looking to monetize their everyday actions. Contributors are extensively vetted and also missions pass through our multi-stage quality assurance pipeline.
The output is a structured data set with the annotations you or model needs. We ship the range of diverse collections from Japanese patient doctor conversations to tasks specific hand-traced egosensic videos.
If you need a data set, browser extensive off-the-shelf catalog or a quest to custom collection. We provide samples for every data set. You review them and if it fits, we license immediately.
CloseThis is just a start. We're powering the future of superintelligence one data set at a time. We do set at a time.
HookCajal (YC W26) is massively scaling formal verification to accelerate scientific discovery.
ProblemLarge language models are probabilistic systems that cannot guarantee correctness. This is especially problematic in mission-critical domains that require mathematical certainty. Formal verification provides exactly this: mathematical proof that code is correct for every possible input, not just the ones you test.
How It WorksCajal is using AI to massively scale this technology - deploying agents to autonomously discover and formalize tools across applied science, beginning with quantum computing and finance.
Every result is machine-verified by Lean's type-checking kernel, guaranteeing correctness.
Congrats on the launch, Luke Johnston and Pedro Nobre!
https://lnkd.in/gqirmPiu
HookMathematics is the language of the Universe, and all modern society was built on top of it. We start a how with the go of solving extremely hard and important problems, problems that humans cannot solve by themselves.
ProblemAnd we try to think what was preventing a AI from solving these problems. While modern elements are powerful, they are emitted by their stochastic nature, as they are fundamentally unable to verify the quality of their outputs.
SolutionWe found a solution to this problem in the lean. In 2013, Leonardo D'Mora created lean and open source theorem Prover. By 2026, its math library Mathlib had surpassed two million lines of formally verified code, helping mathematicians solve the hardest research problems.
Today we see that we are right. Current AI systems, when grounded in formal verified mathematics, with tools like lean, are already outperforming humans in the IMO, international math, all-impid. And we are even starting to see people with the help of AI solving sentry old math conjectures that have remained unsolved for hundreds of years.
Lean has been crucial for this progress, providing an objective source of truth for AI models to work with. At Kahall, we are building infrastructure to massively scale this work, expanding progress from mathematical theory to applied domains with genuine real world impact.
ProblemQuantum hardware is scaling fast. In five years, cubic counts have grown 26-fold. The bottleneck is now algorithms.
Kahall uses formal verification to discover quantum algorithms with proven correctness and speed up and applies the same approach to classical algorithms in finance. Our other results in one of motion theory and finance are promising.
There is still a lot of work to be done. And most of applying mathematics haven't been formalized so far. We have begun to see evidence of a scaling rule with our approach.
As more tools across applied mathematics are discovered than formalized, the systems progress compounds, accelerating future discoveries. We can only begin to imagine how the world will look like a year from now.
CloseLet alone two three five or ten years from now. One thanks for sure. It's a great time to be alive.
HookGrade (YC W26) is building payroll for performance, helping companies pay their workers based on results.
ProblemToday, about $50T a year goes to worker pay worldwide (half of global GDP!). Payroll systems still pay people for hours worked, but hours do not match real output. Even if you pay on performance, it still lives in spreadsheets, screenshots, and manual approvals.
SolutionGrade fixes this by making performance pay run like real payroll. You set the pay rules, they track results, and run payouts in bulk.
They started with creator payments because companies were stuck with a bad choice. Pay a flat fee up front and take the risk, or try to add bonuses and end up with clunky spreadsheets.
TractionIn the last 30 days, companies used Grade to pay out $380k+ to creators, up 120% MoM.
Congrats on the launch, Lotanna Ezeike 💳 and James Heaney!
https://lnkd.in/gFCTx3Rp
HookWhich piece of content would you pay the most for? If you pick this one, you're wrong. They all go pay the same.
Context$650 billion is spent every year on digital advertising. And 40% of it goes down the drain because of inefficiencies like this. The gambling needs to stop.
So what have spent actually matched performance? What was great? We can show you.
How It WorksHere's how it works. First, set your rate. Then we track performance automatically.
TractionAnd with one click, you could pay any creator, anywhere in the world with any currency or payment method. If you could measure for outcome, you should be able to pay for it. And it's already happening.
Think sales and affiliates with commissions. Recruits us with finders fees because the more you perform, the more you earn. And we see this expanding to every industry in the world.
Now that's a $50 trillion problem every year. That's why we're building the payroll system for the future of work. If you're marketing with creators, it's time to pay for performance.
And join brands already serving 185 million customers worldwide. The future belongs to those people for.
HookLegalOS (YC W26) is the AI-native immigration law firm that delivers visa petitions in as little as 48 hours, not months—pairing AI with highly experienced immigration attorneys.
ProblemImmigration law is broken for startups. Traditional law firms take 8 weeks to prepare a petition, charge up to $35k per visa, and give customers zero visibility until it's too late.
TeamQuality is inconsistent- leading to RFEs, delays, and denials. The stakes are brutal: one denial can force a founder to leave the country or cost you a critical engineer. Startups can't afford this.
TractionLegalOS uses specialized AI agents built on 12,000 winning petitions. Every filing is reviewed by an attorney with 40+ years of experience, and delivered in as little as 48 hours—not weeks. They handle O-1, EB-1A, EB-2 NIW, H-1B, L-1, TN, and have filed dozens of applications with a 100% approval rate so far.
Co-founders Rachel Asir and Matthew Asir grew up in their family's immigration law firm, watching thousands of cases and learning what makes petitions win or fail. Claire J.(ex-TikTok) built the AI infrastructure. They're solving a problem they've seen up close for decades.
Congrats Rachel Asir, Matthew Asir, and Claire J. on the launch!
https://lnkd.in/eqEB3RgA
HookWhat if your immigration case moved as fast as you do? We're legal OS, the AI powered immigration law firm that can turn around petitions in 48 hours. In the best part, we are immigration experts, 40 plus years of experience, and we have analyzed 12,000 cases, so our applicants have 100% approval rate.
SolutionIf you're a founder on a visa, you know the drill. 15,000 dollars, a tainer, weeks of back and forth of the lawyers, and you still end up doing half the work yourself. That's exactly why we've built legal OS.
So if you are an employee, you need a visa fast, we're here to help. Check us out today.
HookReturn Signals (YC W26) helps e-commerce brands decrease returns and increase LTV with proactive high-touch support.
Over $1T in apparel, footwear, and accessories is sold online. $300B of that gets returned. Many of these returns are driven by a poor post-purchase customer experience.
SolutionReturn Signals' AI proactively checks in with customers via SMS when their items are delivered and intelligently recommends appropriate exchanges when customers are unsatisfied. These texts convert 70% of return intents into exchanges, improving absolute margin by 5-10% while increasing CSAT and LTV. All this without increasing overall return rate.
CTABook a demo now at www.returnsignals.com
Congrats on the launch Alejandro Zaniolo and Ilya Valmianski!
https://lnkd.in/gM6ktkht
HookOver a $3 billion of a parallel foodware jewelry and accessories is so online. People are not visit in physical stores, they are not getting great service and they are not building brand loyalty. My family has run two parallel brands that sells both online and incandes to our stores.
SolutionYet, the online experience does not match what we offer in person. Over $300 billion of online fashion sales are return every year. These hearts' brads, but it doesn't have to.
Because in encolmers, the journey doesn't stop at checkout, but that is exactly what most brands do. That's why we built return signals. We believe AI can give every customer the kind of high-touch support that used to be reserved for VIP clients only.
TractionWe go beyond traditional support. We practically check in when the item arrives and if something does go wrong, we suggest the best exchange option tailored to every customer and designed to drive attention.
For brands, we are seamless to integrate and offer comprehensive reporting as well as the ability to ask any question and have our AI read through thousands of relevant conversations and share insights. For existing clients, 70% of their returns become exchanges.
That means instead of losing all that revenue, they get to keep it. The brands used to have the shoes which in scale and care. Now they don't.
We don't really need to do a customer to continue after a tick-out. And now they're there.
HookEnterprises have gone all-in on AI. But most have no idea what tools their employees are actually using, what they're spending, or where sensitive data is going.
Oximy (YC W26) fixes that. It sits on your network and give you complete visibility into AI across your organization. Discovery, spend intelligence, and data protection in one platform. No endpoint agents, no manual tracking, no waiting for something to go wrong.
TractionOximy is already processing millions of requests daily for enterprises in financial services and healthcare. If you're a CISO, CTO, or IT leader trying to get a handle on AI sprawl, check them out at oximy.com.
Congrats on the launch Naman Ambavi!
https://lnkd.in/gCAH6nxY
HookSo, you spend billions of dollars in air and now you have 47 different fuels to wealth of inventors and zero-one rents. And your company's confidential data is all over the place. Hi, I'm Naman and give the docs me to solve this exact same problem.
SolutionOxity helps you understand what your employer is doing with air. Which clones are they using? How much they cost?
And, where you're the idealist room? Here's how that looks in practice. Let's say, you marked in game spending thousands of dollars on similar air designs.
We can help you identify and cut those costs. It can also show you which developers are getting the most out of air coding games. Human outdoors learning can secure that clean book up your data on an image.
TractionWhen someone from your sales team tries to upload sense of the customer information it when you pay a赤, while singing can help you block him. The way we do this is we sit on your firewall because every day I add that it never exists.
We communicate while in it. Today, we are working with enterprises and financial services, growth stage companies and healthcare, processing millions of requests every day across thousands of different things.
SolutionJust you will have to choose between speed and safe. We are building a layout that helps you see yesterday I should and still sleep well at night.
HookPanta (YC W26) is an AI-native commercial insurance brokerage.
Their agents work with real file systems, log into carrier portals, fill out applications, email underwriters, negotiate terms, and follow up on the phone — across thousands of clients at once. They've been in production since December.
ProblemGetting a complex business insured takes ~50 steps over two weeks. 95% is coordination, not judgment. Every broker maxes out around 400 clients because no human brain can hold the real-time state of that many cases. The team spent months selling AI tools to brokers before realizing the bottleneck was never any single step — it was the human holding it all together.
TractionSo they stopped selling software and became the broker. In fact, an explosives manufacturer came to them after being turned down everywhere — 70 pages of forms, 100+ emails to carriers. Their agents placed it in three days.
TeamThe founders of Panta are both licensed brokers who've placed hazmat trucking, steel mills, hazardous gas facilities — hundreds of placements by hand. Everything Panta automates is a workflow they've personally executed. Vincent and Frank are friends since high school, building together in SF seven days a week. Vincent previously built Vertex AI's recommendation models and NotebookLM at Google. Frank was a senior engineer at Apple building their Gen-AI chatbot and is a top-10 contributor to the Rust and Gleam ecosystems.
Congrats on the launch, Jiangda Wang and Vincent C.!
https://lnkd.in/g_aTn3RV
HookI'm Vincent, co-founder of Panta, which is an AI Native commercial insurance broker. We get hard to ensure business covered faster than anyone in the world. Processed to get commercial insurance sucks right now.
ContextFirst you call broker and they promise to give you a quote in three days. Then they go silent for seven days and more than half the time they might just straight up go steal. They make fake promises they cannot keep and they have zero incentives to move fast.
TractionThis is a $1 trillion industry running on the tag from the 80s. We can turn it into an efficient marketplace with open-quality AI operators running on MacMainees. Our AI agent redequments, parse emails, and handle calls for calls thousands of clients.
They manage the pipeline and chase the anorider to get the policy bound. Then, in general, certificate in just minutes instead of days. At Panta, where combining licensed commercial insurance broker with AI agents to create next gen of infinitely scalable brokerage, she happens.
HookPolymorph (YC W26) helps consumer and self-serve teams deliver Duolingo-quality lifecycle messaging faster and better than traditional analytics can.
ProblemMost teams already have the data, but messaging is still guesswork across different users and channels like push, in-app, email, and SMS.
How It WorksPolymorph turns product behavior into live user profiles, then automatically personalizes timing, channel, and content for each user, with clear attribution on what drove conversion. One early customer already saw a 3.6x lift in reactivation.
The team is using their experience being in engineering leadership at Meta Ads, scaling ML infrastructure at Scale AI, and building quickly and safely at startups like Gusto, Opal, and Nira Energy since they want delightful experiences for every product.
usepolymorph.com
Congrats on the launch, Andrew Sy, David Nie, and Manas Purohit!
https://lnkd.in/g_-p_qjX
HookEvery user is telling you something. You're just not listening. Consumer and self-serve teams look like this.
SolutionDrowning in endless dashboards, mountains of data, infinite queries. Well, say no more. Introducing Polymorph.
TractionYour automated growth engine. We process billions of signals in real time across your entire stack through one click integrations to tools you already use to build living profiles of each user that continuously learns from product usage, support, and revenue signals.
And automatically personalize their experience, starting with notifications. So your customers on board faster, deepen their engagement, and keep coming back. Your users deserve to feel special.
HookDidit (YC W26) is building the fastest way to verify real humans online.
SolutionAs AI makes it easy to fake faces, voices, and videos, proving who someone really is has become a core problem for the internet. Most identity solutions today are slow, fragmented, and built on legacy systems that hurt onboarding while still letting fraud through.
Didit rebuilt identity verification from scratch. Their all-in-one platform unifies ID verification, biometrics, liveness, authentication, and fraud prevention into a single system optimized for speed and conversion.
TractionReal users verify in seconds, bots and fake accounts are blocked by default, and teams no longer need to stitch together multiple vendors. With simple pay-per-use pricing and integrations in under a few hours, Didit is already used by hundreds of B2B customers.
Congrats on the launch Alberto Rosas and Alejandro Rosas!
https://lnkd.in/dGQYgdAb
What you like here?" The world has been defeated by Bob Bryan Retreats a storm
HookVOYGR (YC W26) delivers comprehensive, up-to-date information about places and local businesses for AI apps, agents, and analytics. Companies already use VOYGR to continuously validate and enrich their place data.
ProblemReal-world interactions run on maps: place data powers millions of apps, and roughly a third of consumer search and LLM queries. But it’s also where things break – locations go stale, attributes stay shallow, and matching is hard.
SolutionVOYGR makes place data accurate and fresh at scale: confirming what’s live, resolving inconsistencies, and adding rich real-world context from web, social and trusted sources.
TractionVlad knows this space inside and out – he worked on the Google Maps APIs GTM, plus firsthand customer experience from building in ridesharing and travel. Yarik has spent over a decade leading ML/search teams at Apple, Google, and Meta.
Check them out at voygr.tech
Congrats on the launch Vlad B. and Yarik Markov!
https://lnkd.in/eMaeAktR
HookWe've always relied on maps. We built the world on them. But the real world changes faster.
ContextMaps can't keep up. Maps work well until they don't have enough information. So a bad map.
Tells a bad story. Maps got smarter, but the world got messier. Places are more than a few tags.
SolutionIntroducing Voyager. Validate and enrich your playstate at scale. Confirm a location as active.
CTACatch discrepancies. Add new attributes from web, social, and trusted sources. Book a demo now at Voyager.tick.
HookInviscid AI (YC W26) AI simulates how air, heat, and energy flow in real time, 1000x faster than traditional solvers.
ProblemData centers are the AI economy's biggest infrastructure bottleneck. Despite this, dozens of projects were canceled last year alone, driven overwhelmingly by energy concerns. The infrastructure that can prove it's efficient is the only infrastructure that gets built. And the problem extends to cold storage facilities, commercial buildings, and many more.
Buildings consume 40% of all energy in the US and EU, with HVAC alone responsible for nearly half of that. Today, this infrastructure is managed using traditional physics simulations that take hours or days to run a single scenario. By the time you have an answer, conditions have already changed. So operators guess, over-provision, and leave massive efficiency on the table. There has never been a way to use physics simulation in real time to actually control what's happening inside a facility.
SolutionInviscid AI changes that. Using advances in state-of-the-art physics-based foundation models and GPU accelerated simulations, they build live digital twins that simulate airflow, thermal, and energy dynamics in real time. So operators can actually see and optimize what's happening, not guess. That means energy savings up to 30%, design cycles compressed from weeks to hours, and real-time control of live operations across data centers, cold storage, commercial buildings, and factories.
TractionThey are actively partnering with data center operators, building owners, cold storage facilities, and hardware engineering teams. If you are dealing with cooling optimization, thermal challenges, or high energy costs, they'd love to work together. You can book a meeting with them at: https://lnkd.in/gjAQypDA
kabir@inviscidai.com | www.inviscidai.com
Congrats on the launch, Kabir Jain and Qiu Ziming!
https://lnkd.in/gdKPqJ76
HookIt takes 60 seconds for the real failure to occur, but it takes up to 60 hours to simulate it. Every get a center, factory and all sorts of facilities depend on airflow and cooling. Despite this, engineers simulate the thermal design once during construction and never again.
ProblemAfter that, it's got fuel or rule of thumb. Meanwhile, real world conditions change every minute. People walk around and airflow shifts, and when something goes wrong, traditional simulations are too slow to predict or prevent anything.
SolutionWe built in this AI to fix this. We created a real-time physics-based digital trend of the facility. We have a numerical GPU-excellited solution that runs full-after-low-enheat simulations in real-time.
TractionWe also have foundational physics-informed AI models. They have state-of-the-art accuracy when evaluated on the diverse terms of equations. When a thousand times faster than the accuracy simulation.
We patch failures before they happen. If there's thousands of designs, in the time it used to take to run one, we are building a real-time physics-informed sphere for data centers full-freakest lithes, factories, and EV batteries.
CTAAnywhere, with thermal performance, really matters. If this is something useful for you, check us out at visit AI.com.
HookVector Legal (YC W26) is building an AI-native law firm for startups, combining experienced startup legal professionals + proprietary AI to help founders close deals quickly and affordably.
The team supports clients from Seed through Series B stages, handling financings, IP protection, customer contracts, hiring, founder transitions, and more.
How It WorksTheir AI platform, VectorOS™, connects seamlessly into existing client workflows (Slack, Claude Cowork, email) so your lawyer is always in the loop, whenever you need.
In addition, the platform equips clients with AI agents to help:
- manage equity and dilution,
- draft documents,
- review contracts with AI,
- maintain investor-ready data rooms, and
- ask questions across their legal docs.
CTAIf you’re a U.S. startup looking for a modern law firm experience, learn more and book a demo at https://vectorlegal.com
Congrats on the launch, Keenan Venuti and Mitch J. Duncombe!
https://lnkd.in/gk3akf6x
HookHi, I'm Mitch. And I'm Keenan. We're building vector legal.
SolutionToday, founders have to choose between either a legacy law firm, that can be slow and more expensive than they like, or handling their legal themselves on an untrusted off the shelf AI system. So we built vector legal, an AI native law firm to equip old clients and council with state-of-the-art legal tech tools, helping your team, or in faster and smoother.
ProblemLegacy law firms today are buried in manual processes that slow things down and increase costs for clients. On the flip side, AI systems make mistakes. They don't know context about your business that your lawyer should, and your conversations with them are not privileged.
SolutionVector legal bridges the gap. You have an AI system built specifically for startups, and whenever you need it, a lawyer in the loop answering questions, getting deals over the line, and being your advocate.
How It WorksWe're excited to demo vector legal's AI platform, Vector OS. Our smart data room automatically organizes your legal docs and extracts key clauses, terms, and insights. Next, our suite of health check agents, scan your documents, flagging material issues.
Our assistant knows your data room inside them out. It can perform research, review contracts against precedents, or create entirely new agreements. AI needs a paper trail.
So we've equipped our drafting agents with track changes, comments, and access to your data room. Finally, there are times when you need your lawyer to be involved. Our team is always one click away, ready to help review, advise, or draft.
CTASo whether you're fundraising, signing your first or 50th customer, growing your team, tagging your IP, and so much more, we're here to help. If you want your legal support to move at the same pace as your ambitious startup, visit vector legal.com to get started.
HookFenrock AI (YC W26) is building AI agents for banks to 10x back office functions.
ProblemThese teams still run on a series of disjointed legacy systems. They are drowning in extremely manual work and backlogs, missing deadlines, and risking regulatory fines.
How It WorksFenrock overlays on your existing stack and seamlessly integrates internal policies and SOPs.
SolutionWith Fenrock AI, banks can:
💰 Make a lot more money by processing loans in minutes, not months.
🤩 Offer a stellar customer experience by resolving complaints in seconds.
💪 Investigate 10x fraud and money-laundering cases with bulletproof audit logs, and pass every regulatory exam.
No need to switch systems or migrate data.
TeamFounders Charu Sharma and Michael M. were one of the first teams to ever build AI agents, and they have deep experience building in highly regulated spaces.
TractionCharu founded one of the largest healthcare API companies, backed by General Catalyst's CEO, and scaled it to over 6 million patients and 100+ employees. Michael built Apple’s first privacy-preserving machine learning at scale, used by billions worldwide, and invented techniques to train ML models without private data.
SolutionIf you’re running a bank, Fenrock AI is a no-brainer. Email charu@fenrock.ai today.
Congrats to the team on the launch!
🚀 https://lnkd.in/gR8sFTa8
HookIntroducing Fenrock, AI agents for back office at banks. In this example, I'm an analyst on the anti-mony-laundering team. Fenrock will help me investigate my case at least 10 times faster.
SolutionI'm an analyst that transaction history of hilltop supply for the last 90 days to see if there's anything fishy. Fenrock connects to my bank core and does all the data gathering for me automatically. All I have to do is review it.
TractionIt explains the customer profile shows me that what's their usual transaction behavior and that there were two unusual transactions. It's showing me the two spikes visually and then in a table format that I can see if needed.
As I continue to investigate this case, I can dig deeper or ask Fenrock to look at the last 12 months. This whole thing would have taken me 30 to 60 minutes. But with Fenrock, I got here in seconds.
SolutionI no longer need to search through multiple clunky systems of the internet. Fenrock just does the work for me and I review and verify.
Hookritivel (YC W26) is an AI-native workspace that helps pharma regulatory teams draft FDA submissions 12x faster.
ProblemEvery month a drug is delayed costs pharma companies ~$45M in lost revenue and more importantly, delays treatment for thousands of patients who need it. Regulatory teams spend weeks manually synthesizing thousands of pages of clinical data, hunting through SharePoint folders for the right documents, and drafting submissions like CTDs, CSRs, INDs, and BLAs.
How It WorksRitivel deploys AI agents that generate comprehensive first drafts directly in Microsoft Word, following FDA formatting requirements and pulling from existing clinical data. Every data point includes citations linked back to the original source. Connected to SharePoint, Veeva, and Outlook, their agents automatically source documents and handle follow-ups, so teams can focus on review rather than administrative overhead.
TeamThe founders previously built AI copilots at Microsoft Research and spent 50+ conversations with pharma professionals to understand the regulatory workflow bottleneck before building Ritivel.
Congrats on the launch Pavan Kalyan, Gunin Gupta, and Nirmit Arora!
https://lnkd.in/gJnVGp47
HookRIDAVAL helps regulatory teams submit FDA documents 12 times faster. Regulatory submissions take months. Thousands of pages of clinical data and listing through SharePoint constant follow-up emails.
RIDAVAL Our AI agents do the heavy lifting for you. First, our agent automatically pulls the right documents from SharePoint and Viva, no more folder hunting. RIDAVAL can draft CSR, CTD, NDA, I&D, and BLA.
How It WorksThen it generates a compliant first draft directly in word, in minutes, not weeks. And when documents are missing, our agent sends follow-up reminders automatically, so nothing falls through the cracks.
Every data point is cited back to the original source. Click any reference to verify instantly. This prevents hallucinations and builds trust.
HookUnisson (YC W26) is an AI-native platform with agents that work as hands-on product experts alongside your Implementation, CS, and Sales Engineering teams.
How It WorksToday, enterprise deployments require specific customer context, hands-on help, and deep product expertise that is hard to scale. Over time, this work gets passed to a small number of product experts, slowing down time to value and turning cost-to-serve into a bottleneck for growth.
SolutionUnisson fixes this by providing customer-facing teams 24/7 access to an AI subject-matter expert, allowing them to scale and save thousands of hours.
Unisson’s agents learn to use any product in 15 minutes, gather context from meetings, call transcripts, and Slack, and then plan and execute tasks like onboarding, change management, and migrations end-to-end.
CTABook a demo if you're looking at ways to scale your customer-facing teams: https://lnkd.in/gTsrKPac
Congrats on the launch Varun Mathur and Tom Achache!
https://lnkd.in/gg-wp4K3
HookToday we're excited to show you unison. At unison we're building AI agents that help customer success, implementation and forward deploy engineering teams scale. We're already trusted by some of the fastest growing startups and enterprises.
ContextHelping them resolve their hardest, most technical customer support. At start off we've our first agent, Explore. In 15 minutes, Explore discovers hundreds of new features and workflows by exploring your UI and code base.
How It WorksAnd as your product evolves, Explore keeps your knowledge base up to date. Explore is also easily extensible. You can add hundreds of new workflows with built-in evolves and tests all without any new API, all support from engineering.
TractionAnd up next, our agent runner. Runner gather requirements and contacts from all your existing tools and handles your hardest, most technical customer tasks, like onboarding, product administration and custom integrations and to app.
How It WorksRunner works with you to build a plan before taking on tasks and knows when to ask clarifying questions. And for product engineering teams, our new runner API supercharges your own agents to give them the skills to use your product like an expert.
TractionIf you're a fast moving company and looking at ways to scale your customer teams, we'd love to show you a demo.
Founder at Menza (YC W26), prev ShortlyAI (acquired by Jasper)
HookYour greatest revenue opportunity is hiding in plain sight. Consumer brands miss out on millions every day, not because of the lack of data, but because the right insight came at the wrong time. Today that changes.
ContextWe help consumer brands see what they need to know when they need to know it. We have 24-7 data monitoring. This is what an insight looks like.
ProblemWe can click through to see more details on this insight, whether it is a misrevened opportunity, wasted spend, an interesting trend, or an interesting anomaly. In this instance, it's a revenue opportunity.
We can ask menzer to dig deeper into this insight. For example, let's say we have a meeting in 10 minutes. We can ask menzer, which products should we be promoting in this campaign?
Just so that through, now menzer will get the answer for us. On the back of this, we can ask menzer to perform this analysis every day at 6 a.m. and send it to our inbox as a report.
Perfect. I'm looking forward to getting this every morning. As someone running a team, I like to get an overview of what's going on in the business.
How It WorksLet's spin up a dashboard from all our important data sources, telling you what I need to know. This includes our ads platform, our social media, our support, and our account and software. Where ahead of you, I put this together while you're talking to them.
Less, our vision is simple. A world where operators do not chase data because the right insights finds them first. The question isn't whether your breakthrough insight exists, it's whether you'll find it.
HookThe ton of his vehicles are already on public roads, where houses are run by robots today, and drones are delivering goods to your door. Both nearly zero historical claims data ensures cannot properly price autonomy.
ProblemInsurers may see only one claim per million miles. But you don't need to wait for claims to understand risk. We build probabilistic models of real-world environments, then run billions of simulations to estimate claims before they happen.
SolutionVALGO is the foundational layer that gives insurance providers the loss estimates to properly price autonomy risk. Rebuilding VALGO, the risk quantification platform to ensure physical AI.
HookDitto Bio (YC W26) discovers drugs for autoimmune diseases by harnessing parasite biology.
How It WorksViruses, ticks, and worms have experimented over millions of years to create proteins that control the human immune system. They use AI to identify these natural, drug-like proteins and engineer them into next-generation autoimmune therapies.
SolutionSince founding seven months ago, Ditto Bio has analyzed 1M+ proteins across primate-infecting viruses, ticks, and parasitic worms. They've discovered thousands of proteins predicted to hit clinically validated human targets. Early experiments show that they have surfaced an untapped source of new drugs that act on high-value targets for immune diseases. Their proteins can achieve extremely tight binding (1–2 nM), among other drug-like properties.
TractionThe opportunity is vast: >98% of viral and parasite proteins remain uncharacterized. Ditto Bio is systematically mining this unexplored biology at scale to find and deploy nature’s best drugs.
Congrats on the launch Adair Borges, Dennis Sun, and Emily Weiss!
https://lnkd.in/gh7fVWsE
HookHi, I'm Bader. I'm Dennis. And I'm Emily.
SolutionAnd we're Ditto Bio. We discover drugs for autoimmune disease by harnessing parasite biology. Autoimmune diseases like arthritis, exima, and inflammatory bowel disease affect one and ten people worldwide.
In these conditions, the immune system, which is meant to protect us, mistakenly attacks the body, causing chronic inflammation and lifelong illness. Despite billions invested in nutrients, nearly 90% of autoimmune drug candidates fail in clinical trials because they aren't safe or effective enough.
TractionA Ditto Bio, we're taking a different approach, learning from biology itself. For millions of years, viruses, ticks and worms have evolved to precisely modulate the human immune system, calming it, redirecting it, and preventing harmful inflammation.
We've discovered thousands of parasite molecules capable of modulating the human immune system. Each one of them could be a drug for patients with autoimmune disease. The best drugs come from nature.
CloseI was in pick from Helom monster venom, Penisol and from mold, and Botox from a deadly bacteria. We're making parasites the next source of transformative blockbuster drugs.
HookOne of the biggest challenges with manufacturing hardware at scale is procurement logistics. Most of the work is painstaking we've done by sifting through endless spreadsheets and emails to find buried information about where your part is.
ProblemWhen they're going to arrive in what's blocking your building. Most hardware, EOP and PLM systems contain information, but they don't do anything actionable. Proofs are deeply buried in cognitive systems.
Their part is suppliers taking an entire beta reply. And the right stakeholders are never in the right room at the right time. In the age of AI, hardware procurement is still stuck in the stalling ages.
SolutionThat's why we're building reframed. Reframed is applied from the bottom of its procurement. It's super simple.
How It WorksYou upload your BOMM into our system, connect your email, connect Slack, and reframed into the rest. It'll earn clear suppliers are which parts are suitable for that. Increase actionable and automated next steps.
This includes sourcing the parts, getting quotes, placing orders, and resolving delays. With each interaction, reframed gets more context. Learning things like tone, and creating waterfalls schedules to back up a lead times and input things.
Procurement is a full-time job. It's extremely manual and expensive in terms of time and capital at any scale. At reframed, our value proposition is simple.
You design the product, we'll go get the components. Book a call with us and start automating your hardware procurement today.
HookSyntropy (YC W26) is building the next generation of agentic coding tools for long-horizon tasks.
How It WorksAs AI coding tools improve week over week, the upper bound looks great—but the practical workflow still demands heavy user orchestration. For most teams, “autonomous” still means hovering over a terminal for hours.
Syntropy completely automates this experience. Co-write a spec with the agent, then Syntropy executes end-to-end in the background: PRD generation with verification loops, phased task breakdown, parallel execution in isolated worktrees, testing, and merging everything back together — go out for lunch and come back to a production-ready PR. Syntropy also joins your Slack to send updates right as tasks complete, and fully supports custom MCP integrations.
CTAPublic access is opening now for teams building real products in real repos! Get started today at syntropy.io, or reach out at founders@syntropy.io.
Congrats on the launch, Saahil Sundaresan and Andrew Kuik!
🚀 https://lnkd.in/gwYVPPzB
HookCoding agents can build anything. Well, not quite, but you know the drill. You just need the right detailed prompting.
ContextThe right hooks, tools, context files, aralization controls, verification loops, and custom sub agents. Hey, you're keeping up so far? Great!
Oh, wait. Subteams are a thing now, and open claw brings new opportunities. And heartaches to the nose are now the new.
SolutionMaybe all this manual setup is a lot. Introducing Centropy. Same models, same capabilities, zero configuration.
Request a feature, collaborate for five minutes. Didn't think of a detail? No worries, we'll catch it for you.
When you're ready, press build. The system optimizes everything else. Then, go for a coffee break.
Take a meeting, or don't. It's your time, after all. Centropy.
CloseStop configuring. Start building. Try now at centropy.io.
HookTerminal Use (YC W26) is hosting infrastructure for background agents that makes harness engineering easy.
ProblemThe best agent applications don’t win on the model, they win on the harness. Creating that harness is painful: endless prompt tweaking, guardrails, and eval curation.
How It WorksWith Terminal Use, host agents built on Claude Agent SDK, Codex SDK, or your own framework. Trigger via our SDK, or on a schedule.
SolutionThe platform is CLI-first, which allows your coding agents like Claude Code & Codex help you debug, hypothesize improvements, and test them.
TractionServe businesses out of the box: isolate data and compute per customer, and permission resources granularly.
Congrats on the launch Vivek Raja, Filip Balucha, and Stavros Filosidis!
https://lnkd.in/gR4_B2bh
HookAI agents are getting remarkably capable. Soon, they'll have the ability to run complex tasks all by themselves. But today, it's difficult to ensure reliability.
ContextIt takes ages to optimize it for your own use case. You spend weeks, tweaking prompts, you both turn guardrails into up-of-guardrails, and there's no sustainable way to manage and scale multiple agents.
SolutionThis is where we come in. Termal use is a platform for building, deploying, and iterating on your back-harn agents. It's designed to be simple.
How It WorksIn minutes, not ours, you can deploy agents to the cloud. What makes Terminal use different comes down to three core principles. Let's say you want to run multiple experiments to test different methodologies on the same outcome.
You can have various agents working the same task under numerous iterations simultaneously. And then, you can have other agents working top of that to determine the best outcome. This saves you tremendous amount of time while iterating.
Our platform is extremely agent-friendly. We make it possible for your coding agents to deploy other agents, communicate with them, access their tasks, and countless other things, all by using just the terminal.
SolutionAnd lastly, you're not locked into our platform. You can use any agent framework you'd like, we don't limit what you can build. Also, everything your agent produces belongs to you.
CTAWe make it easy to sync all data with your own source systems. Building and improving back-harn agents has never been easier, and we can't wait for you to try it.
HookSo scout out handles construction back office, but first I want you to imagine your construction company owner who just got the job set at 6 p.m. and still has estimates to draft jobs to track and proposals to set.
DemoI don't want to deal with any of that. So let me show you how we can actually get this done. First thing I have building plans I need to measure now.
How It WorksNormally I'm gonna do this by a hand or a piece of linen house, but here I can just upload the plans, calibrate the scale, and start clicking. Scat out is gonna lift through my plans, measure it for me, area layer, feet, individual counts of items, take-a-goster done.
SolutionNow I can take all those measurements and plug them into my estimate and scout out was gonna format all of it into a legit proposal. I can send straight to the client for signature. So no more copypacing into Word or whatever, everything's in one place ready to go.
CTAIt's done in a matter of seconds. Let's go to the AI and go check it out. Good job.
HookCompresr (YC W26) helps you boost context management with 100x compression.
ProblemHave you ever struggled to fit all relevant context into the model’s window, only to see the model failing to grasp it? Waited for Claude Code to finish yet another 3-minute compaction session? Got surprised with the OpenClaw bills?
How It WorksCompresr is on a mission to make you forget about managing long model/agent’s context. With their API, you can reduce long context to the essentials the model needs for any given request – making its generation better, faster, and cheaper.
Compresr cares deeply about seamless and simple integration into agentic workflows. Their open-sourced proxy takes a minute to set-up, and optimizes your Claude Code / OpenClaw context from the moment you finish installation. You will experience no change, apart from better performance and smaller bills.
Congrats on the launch, Ivan Zakazov, Berke Argın, Oussama Gabouj, and Kamel Charaf!
http://compresr.ai/
https://lnkd.in/gmcdt3eB
HookYou might think it's a bug, it's actually context rot. Context rot context rot context rot context rot. Have you ever fed your AI so many documents started hallucinating?
DemoThat's context rot. We are here solving time or compressor. Let's look at a real example.
SolutionWe want to ask a question about 15-pall gram essays. On the left side, we have the essays in their original form. On the right side, we compress them using the question at hand.
TractionA concrete question could be, how to found a successful startup. The model working with the compressed context delivers same quality answers faster while reducing costs. We also enable effortless integration with your agents through our gateway.
Here we ask Cloud Code to build a fully functional chess app. On the left, we are building with raw, clawed code. While on the right, we are using our context gateway on top of it.
With compressor, we have 20% lower cost, 25% less latency, and 44 more passing tests. Everything you need to try out our method, including our SDK and our context gateway, on compressor.au. Make every token count.
HookMendral (YC W26) keeps your CI green and fast by finding what broke, why it broke, and helping ship the fix.
How It WorksTeams lose huge amounts of time to flaky tests, slow pipelines, and recurring build failures. Most tooling tells you what happened. Engineers still have to dig through logs, rerun jobs, and patch brittle CI over and over. With more code being produced than ever, that drag only gets worse.
TractionMendral watches your CI/CD, diagnoses failures, flags flaky tests, finds slowdowns, and opens PRs with concrete fixes. It is already running in production for 15 teams, including PostHog. We are founding engineers from Docker and later co-founded Dagger. We built delivery tooling for years, and we are opinionated about one thing: teams should not need dedicated platform headcount just to keep shipping.
Congrats on the launch Sam Alba & Andrea Luzzardi!
https://lnkd.in/gD5RP3DH
HookHi, I'm Sam, a co-founder of Mangeloll. Mangeloll is an AI DevOps engineer that manages your CI CD. It improves your build, fixes your test, improve the security and performance of your software delivery.
DemoLet me show you a demo. First trial is one of our early customers. Mangeloll is looking at every CI logs, every commits, every configuration changes from first trial and look for opportunities to help improve the speed.
ProblemAny proof of your reliability and security. Let's look at one example with a flaky test. Flaky test are hard to spot because they don't always fail.
DemoAnd so in order to spot that the test is flaky, the agent needs to look at the failure but also at historical data. So here it's studied the flaky test by looking at all the data available and gathered over time.
ProblemIt would accomplish food cause analysis about what the problem happens and also propose remediation options. The team from first hug accepted one of the remediation and the agent jumped in and opened a request.
And then it was merged to the repo. The agent is also available on Slack. It joins your Slack channel, exactly like a human software engineer would do.
And it's not telling you when it finds something to do. You can then jump into the thread and trigger an implementation directly from the Slack app. You can even ask Mangeloll questions like, do I have any flaky test?
TractionAnd Mangeloll will run an analysis for you and search directly on Slack. To them, Mangeloll is deployed in 15 different companies. If you would like to be one of them, we would love to give you a quick demo and unball you in few minutes.
HookPermitting delays don’t have to stall projects — or drain budgets.
AutoSitu (YC W26) reviews site and building plan sets in minutes and surfaces what a city is likely to flag before you submit.
ProblemToday, development review is constrained by manual, repetitive work and inconsistent feedback loops: completeness checks, code/policy cross-referencing, and comment cycles that trigger resubmissions and late redesigns. The result is expensive—developers rack up carrying costs, architecture teams lose margin to unplanned rework, and understaffed city teams carry growing pressure to support housing and development.
SolutionAutoSitu helps teams catch issues earlier, submit cleaner sets, and reduce back-and-forth—making approvals faster and more predictable without adding headcount.
Congrats on the launch Asher Lin & George Zhai!
https://lnkd.in/gXsnhnyd
HookOki-Tax and developers don't lose time in construction, but they lose time in the back and forth of development plan reviews. Typically, 3 to 6 months before a project can even hit the ground. For developers, that's 250k plus carrying costs.
SolutionFor arc text, a single recent middle could wipe out 25k plus profit margin. That's why I would build AutoCity. To help you catch issues early before you first summing to the stage.
Hook10x Science (YC W26) is the AI-native solution to unlock fast, accurate, and scalable protein characterization to accelerate drug development.
How It WorksProtein-based drugs now make up a majority of new therapies and advances in AI are flooding the drug discovery pipeline with more candidates than ever. But the bottleneck isn't candidate design, it’s candidate characterization to identify the best drug for clinical success.
Antibodies, cell therapies, and engineered proteins require molecular-level characterization beyond the limits of existing tools. This results in weeks to months of analysis per candidate, high costs from market delays, and manual workflows that are challenging to scale and reproduce.
Traction10x Science is the unlock. They are building AI-native infrastructure for protein characterization. With 10x Science, analysis times are reduced from weeks to minutes, yielding $150,000+ time savings per team/month. You also get reproducible, enterprise-ready results with the click of a button and without the need for extensive expertise and slow manual curation. You can even scale your pipeline to process more candidates exponentially without linearly scaling your efforts.
TeamThe company was founded by Stanford researchers from Carolyn Bertozzi’s group with backgrounds including expertise in the fields of mass spectrometry-based proteomics and chemical biology with 37+ peer-reviewed publications, development of new methods in structural proteomics for protein folding and epigenetics, and a 2x YC founder who also studied on the cutting edge analytical chemistry and cancer research.
Congrats David S. Roberts, Andrew R., and Vishnu R. Tejus!
https://lnkd.in/dbZDd2XJ
HookWhat are the reasons that drugs fail? According to the FDA, 74% of all drug rejection sites, chemistry, manufacturing and quality control deficiencies. Slow and accurate protein characterization is a major culprit.
ContextAnd it's killing us. Tenic science is the AI native platform that farm companies use to accelerate drug development and unlock critical insights from raw data that are otherwise missed. We have spoken with other researchers and they all feel the same frustration.
ProblemScientists would painstakingly spend days manually creating data and stitching together results from fragmented tools. There's just a clear lack of tools out there to characterize biologics. Memory is now directly integrated into your workflows.
SolutionWhether it be sequencing drugs, profiling molecular changes, or even identifying things you didn't expect could happen, our platform agentically solves that for you. What sets this team apart is that they have lived the problem, literally.
They're not outside observers building tools for other people that they're not familiar with. These are scientists who met each other in my lab who recognized that there was an unmet need in AI software development for protein characterization.
ProblemAnd they just got together and they solved the problem. As a scientist myself, I recognize the quality of the advances behind this platform. It's built with the kind of care and validation that the industry demands.
CTAJoin us at 10x Science where drug development finally keeps pace with discovery. We are 10x Science.
HookSpotPay (YC W26) is building the Global Bank Account — one account that lets anyone send, receive, and spend money locally and across the globe.
SolutionModern financial tools still feel like a privilege, and billions of people remain unbanked. Most cross-border services today focus on building innovation on the sender side, while the receiver side is still stuck on slow, expensive and outdated technology. SpotPay is built to fix that.
How It WorksSpotPay uses modern blockchain technology to give every user a verified account tied to their real identity. That foundation allows them to issue a SpotPay Credit Card and integrate seamlessly with Apple Pay / Google Pay — even for users who don’t have a bank account.
SolutionSpotPay also provides merchants with the ability to accept payments with near-instant settlement. For SMBs, this often means they can utilize revenue from their sales up to 24 hours faster than they would using traditional finance and maintain healthy cash flow as a result.
TractionThey launched just over two months ago, and today they’re already live in 40+ countries with over 42% avg increase in Total Payment Volume week over week.
TeamSpotPay is founded by Zsika Phillip-Baptiste and Thomas Césaré-Herriau, whose backgrounds span Google, Brex, and Stanford GSB, with firsthand experience growing up, living, and working in multiple countries around the world. Their team has the technical depth, product experience, and lived understanding to build the next generation of global financial services.
Congrats to the team on the launch!
🚀 https://lnkd.in/gr25ByWw
HookPayments shouldn't slow you down. Tap your phone to pay with Apple Pay or Google Pay. No need to carry cards or even your wallet.
Freedom of online shopping that actually works with the Spot Pay Calipso card. Send and receive money with friends, locally and securely. Receive money from the US in seconds and spend it right away.
Send and spend across the Caribbean and all over the world. All with one app, Spot Pay. Money, reimagined.
HookVoice AI has enormous potential, but adoption is still limited by two persistent challenges. Conversations often sound robotic, and organizations hesitate to trust AI with customer communication. Many voice systems still behave more like scripted workflows than real conversations.
Samora AI (YC W26) is built to change that. It speaks naturally and in the customer’s own language and dialect. It learns from past conversations, completes tasks end-to-end, and provides full explainability into what the system is doing. When needed, it escalates seamlessly to Samora human agents.
How It WorksCompanies can deploy these agents without investing in their own AI infrastructure, operations teams, or call centers. The product is already being used by organizations across recruiting, financial services, education, real estate, and government services.
TractionIn the last two months alone, Samora has grown 10x. The founding team previously deployed voice agents for national-scale government and UN programs reaching more than 500,000 users across 80+ countries.
Congrats to the team on the launch!
🚀 https://lnkd.in/ghEAddZv
HookNo one likes feeling ignored when they need help. If you're going to automate customer service, your customers deserve excellence. Like picking up immediately.
ContextBuilding maintenance? How can I help you? Hi, I need someone to fix my sink.
ProblemNo problem. I'll have someone there as soon as possible. Handling requests faster than a human.
It's leaking all over I need help fast. Stay with me. I'm getting maintenance on the line now.
TractionAnd smoothly handing off to a human when needed because great service isn't just fast. It's clear, aware, and in the language your customer speaks. That's where some more comes in.
Tell us what you need and we design, deploy, and manage a fully customized voice agent for your business. Your customers get natural, human-sounding conversations in 20 plus languages. With domain aware intelligence ready to serve your customers 24.
How It WorksYour team gets policy-driven automation that understands intent, completes tasks, and escalates seamlessly when needed. No scripts to manage. No AI models to train.
No operational burden. Some more connects directly to your systems. Resolve requests and hands-off smoothly to train some more operators when needed.
You stay in control with full visibility into calls, transcripts, and performance. Your data stays protected with encryption and strict access controls. Go live in as little as one week.
CTASamora, the voice your customers deserve. Visit Samora.ai today.
HookBeeSafe AI (YC W26) stops trust-based scammers before they reach your customers.
ProblemSeemingly harmless messages like “Are we still up for golf?” are the start of a sophisticated social engineering scam, sometimes referred to as “pig butchering”. Scammers strike up a conversation to build trust with their victims over time, eventually convincing them to authorize real-time, irreversible payments via P2P apps, crypto, or wire transfers.
$12B was reported stolen in trust-based scams like this in the US last year and Banks, Telcos, and Government Agencies lack visibility as attackers bypass traditional fraud prevention systems.
TractionBeeSafe AI stops trust-based scammers before they reach your customers. Their system intercepts and engages with attackers to extract intelligence, including financial mule accounts, fraudulent assets, malicious domains, crypto drainers, and other related infrastructure. It also provides real-time, ground-truth data and has already helped organizations identify and intercept scammers pre-transaction to prevent victim losses.
TeamCongrats on the launch Ariana Mirian, PhD, Daniel Spokoyny, and Nikolai Vogler, PhD!
CEO @ Jinba (YC W26) Enterprise Workflow Automation | Ph.D. in CS
HookEvery large company has people doing very repetitive document processing work. It's a grueling job and mistakes are costly. This is where Eigenpal comes in.
SolutionEigenpal lets you automate complex tasks and transforms people into superhuman AI supervisors. Eigenpal employees become massively more productive. For example, Bankback office employees manually process mortgage applications.
One of the crucial steps is verifying salary from a bank statement. Let's describe this to Eigenpal's AI agent, asking it to read any bank statement, identify the employer, and some monthly salaries. Our agent builder replicates this workflow with ease.
How It WorksLet's run our new workflow on a 100-page bank statement. Following the process, it extracts all the monthly salaries in less than a minute and can process any number in parallel. It's important that a business trusts their AI workflows.
TractionTest your automation versus a dataset to understand which tasks are fully automated and which need a human check. In this example, we see that the workflow is at 100% accuracy, so we can deploy with confidence.
How It WorksOnce a workflow is built, let's deploy it as an API or a shareable link. Eigenpal automates all paperwork. Process any document, extract structure data, and apply complex business logic.
CTAFill in templates, trigger APIs, or integrate with existing systems. Validate every result. Try us for free by visiting Eigenpal.com.
HookBooko (YC W26) is bringing dynamic pricing to any business that sells time slots - improving utilization and turning unsold inventory into revenue.
ProblemTime-slot-based businesses all face the same issue: every unsold hour is $0 revenue. Once a time slot goes unused, that revenue is gone forever. Demand is uneven by nature. Peak times book out, while slower hours sit empty. Yet most businesses still rely on static pricing, giving customers no reason to choose those slower times and leaving capacity unused.
How It WorksBooko integrates with existing booking systems to automatically adjust prices and incentives based on historical utilization, real-time availability, and demand. Similar to how airlines or Uber price unused capacity, Booko helps businesses fill more bookings during slow times while staying inside their existing workflows. Early customers see around a 20% revenue lift by filling time that would have otherwise gone unsold.
SolutionBooko works with any business that sells time, from fitness studios and med-spas to tutors, consultants, and professional services.
Congrats on the launch Arjun Saluja and Will Hall!
https://lnkd.in/gYaU2uUE
HookThe year was 2014, and Uber realized something the airline and hotel industries had already known for decades. And it's taken a appointment-based businesses more than 10 years to catch up. It's not rocket science.
ContextNot all hours are equal. Tuesday at 2 p.m. isn't Saturday morning, so customers choose the convenient hour, and when an hour goes on book, it's worth zero.
SolutionThat wasn't free time. It was unsold inventory. Booko fixes that.
TractionPeak hours stay full price, while slow and last-minute slots adjust dynamically. Customers choose the incentive. Empty time gets filled.
The result. Customers across multiple industries see around a 20% increase in revenue. The selling a slow hour for less is always better than letting it go to zero.
Founder Mister ConTech · AI & Automation for Construction · Training & Implementation · Ex-Bulldozair (YC S16, Exit)
HookOkay, let's go ahead and create a new project. I'm gonna call it demo now I can simply upload my specs and drawings and click the create project button now that we're in Let me show you around the platform The first thing that pops up is our drawing and we can also click the specifications tab and look at the specs We uploaded only back to the drawings tab You can annotate it measure and mark it up all within our platform now Let's try and ask it to do an estimate.
DemoI'm gonna ask it to extract the full electrical scope of the project and create an estimate for it First it's gonna go through the drawings and specs to figure out the requirements then It's gonna go through our database and figure out what assembly is to use for this project based on the info it gathered from the drawings and specs After that, it's gonna identify labor units for all the work items Finally, it'll scrape the live material costs of all the work items And I'll put a material sheet for us to look at She's curl up. You'll also find that we provide proof of work for everything That's just it shared so if it says it found something on the drawings with the specs It will provide the proof of work for that and show you exactly where it found it We can see the full material sheet in the estimate tab at the top We can edit our labor rate overhead profit margin and material tax and we can even export it as a CSV or PDF And there you have instead of spending hours during yourself.
You now get a full material sheet in a few short minutes
HookPayna | YC W26 helps companies with file lending, debt collection, insurance, mortgage, and money transmission licenses across all 50 states in record time, then automates renewals and ongoing regulatory compliance so teams can focus on growth and real revenue.
ProblemA huge swath of traditional industries, together with a new wave of digital asset, stablecoin, and crypto companies, is running into the same fragmented 50-state licensing landscape, each with its own forms, portals, evidence requirements, and renewal cycles. Together, these sectors represent trillions of dollars of economic activity, yet the infrastructure for licensing and compliance still runs on spreadsheets, inboxes, outside counsel, and legacy government systems like NMLS and NIPR.
How It WorksPayna turns that fragmented process into software. They structure licensing workflows, map regulatory requirements across all 50 states, organize evidence and filings, track renewals, monitor rule changes, and integrate directly into existing company workflows.
TractionPayna is already live with companies in debt collection and digital assets operating compliantly on tens of millions of dollars each month, and they just signed six-figures in annual contracts shortly after launch.
CTAIf you know founders, compliance leaders, general counsel, or operators at companies looking to file and maintain licenses, reach out at hello@payna.com.
HookFullSeam is your AI-powered finance and accounting teammate.
SolutionFullSeam agents log into a company's existing accounting tools and perform routine tasks that accounts receivable and accounts payable teams usually do by hand - like communicating with vendors and customers, recording payments, and updating financial records.
The founders are the same team that founded TaxProper, their first YC company. At TaxProper, they paid more than $1 billion a year in property tax payments for customers before being successfully acquired by Opendoor.
TractionAt TaxProper, they were growing quickly — but their accounting team was drowning in invoicing, reconciliations, customer questions, and exception handling. The software they used kept records, but it didn’t do the work. Humans still had to chase emails, upload documents, correct entries, and follow up until payment landed.
How It WorksAccounting teams don’t need another dashboard — they need teammates.
Congrats on the launch Thomas Dowling, Aaron Coppa, and Geoff Segal!
https://lnkd.in/g8TxXvNp
[♪amız plays, amp axe extremely low solo Anna cast track Starting theme plays
HookWideframe (YC W26) is an AI agent that speeds up the 75% of video work that happens outside the Non-Linear Editor.
SolutionIn the first 75 days since launch, Wideframe has onboarded more than 50 brands and agencies. Many of them make tens or hundreds of videos per week for ads or organic social. 75% of their video work happens outside of the NLE and Wideframe helps accelerate this work to save teams time and allow them to do more with their videos.
TractionUsers link their video libraries and then work with a context-aware agent to tackle search, editing, and generative tasks. The high-performance app lives on the machine and works across the filesystem. It can even natively handle Adobe Premiere Pro (.prproj) files in and out.
Congrats on the launch, Daniel Pearson and Zachary Kim!
https://lnkd.in/gtgvU69c
HookSurrender, growing about AI replacing. Editors are going to be replaced. Companies are quickly adopting AI.
ContextCreatives are done. They're wrong. Meet Wide Frame.
SolutionThe truth is, 75% of video work happens before the real edit begins. Wide Frame handles the grunt work, searching, scrubbing, organizing, so you can spend more time on the creative parts of editing. Link your footage and wide frame analyzes and indexes everything at superhuman speed.
Wide Frame finds footage by meaning not file names. Ask for wide shots of killy typing and get results in seconds. It organizes bins real way.
Averile, pveral, reactions, cutaways, pulled like a trusted assistant editor. While you put the finishing touches on one video, wide frame is already analyzing, organizing and assembling your next one into a premier ready sequence.
DemoCreatives will spend more time in a flow state versus doing admin work. Now watch how wide frame help us make this video. Ricky shot the footage and linked it.
How It WorksWide Frame started indexing immediately. The worst part of Ricky's job is spending two hours after every shoot, labeling files, organizing into folders and uploading to the cloud. Now it only took him five minutes.
Ricky can use the extra time to focus on making better shots or planning for his next shoot instead of being rushed. When Damien, our editor, opened the project. It gave wide frame a simple prompt.
Instead of 12 hours of scrubbing footage and building project files from scratch, wide frame reviewed every clip and matched it to the brief. Damien took the assembled sequence into premier to add his finishing touches.
With the extra time, he could focus more on the animations, color grading and details, while still delivering this video in a week. Wide Frame doesn't replace editors. It gives them more space to create.
CloseMore space for the craft, more space for the story, more space for the work that actually matters. Try Wide Frame free for seven days at wideframe.com.
HookAvoice (YC W26) is a suite of AI tools that helps architects automate admin work and turn past project experience into a shared knowledge base, saving 15 hours per architect per week.
ProblemArchitects spend 70% of their time on non-design work: writing specs, reviewing submittals, researching codes, and coordinating across teams. This work has to be technically precise, yet it's still managed through clunky legacy software that hasn't changed in decades.
TractionAvoice lets firms write specs in minutes instead of days, review submittals, respond to RFIs, and search across past projects to reuse proven decisions. Firms are already using it on projects worth hundreds of millions of dollars. Avoice helps run the tedious, but important, side of architecture, so architects can get back to what they do best: design.
Congrats on the launch Chawin Asavasaetakul and Chawit Asavasaetakul!
https://lnkd.in/gswFZBvE
HookWelcome to a voice. The AI workspace built for architects. Practice is around the world used a voice to streamline documentation, strengthen coordination, and unlock years of project knowledge, saving uses up to 25 hours every month.
ContextEvery practice works differently. A voice adapts to your workflow, and supports every stage of a project. From schematic design through construction administration, assistant is your intelligent design partner.
SolutionIt reads drawings, interpret schedules, drafts documentation, and handles hundreds of architects specific tasks, from code checks and material research to RFI responses. All powered by context from project files, firm standards, building codes and product databases.
How It WorksWorkflows automatically complex architect workshops that normally take hours of coordination. Run QA across entire drawing sets, batch submittals and RFI's, review contracts in minutes. What wants to update these now to these minutes, projects bring everything together, growing schedules, specs and products in one place.
In one click, a voice cross reference is drawings and schedules. Flag decisions early, and auto-generate specs and schedules aligned to industry standards and your firm's approved materials. Library centralizes your firm's knowledge, past projects, standard details, specs and materials, all searchable, and reusable across your team.
Turning your hard-earned expertise into a living resource. Our voice is the modern workspace built to let your practice do when it does best. Design.
HookSalus (YC W26) is building runtime guardrails to validate your AI agent’s actions before they execute.
SolutionExisting practices like evals, output scoring, and observability are all necessary—but all reactive. Salus offers a proactive solution that inspects actions before they can run.
How It WorksBy adding their API on top of your existing codebase, incorrect actions are automatically blocked and your agent receives actionable feedback to self-correct at runtime.
SolutionActions are approved only if they’re backed by evidence from the agent’s execution and conversation history, comply with your configured policies, and pass built-in PII and safety checks.
How It WorksIf you’re deploying agents, improve their correctness and reduce your reliability workload with Salus.
Congrats on the launch Kevin Pan and Vedant Singh!
https://www.usesalus.ai/
https://lnkd.in/gRwjjggW
HookHi everyone, I'm Kevin, co-founder of Sales. It's 26 everyone wants to use AI agents, but they still make mistakes that rates that make production of web-in-grace. As sales we've fixed that.
DemoBy importance to feelings of code, we validate every action your agent takes before it executes. We've blocked the bad ones and we give your agent the context of needs of the self-cred. Let me walk you through a quick demo.
One common use case we've seen is customer service agents. Here we're running customer service tasks for an airline from the task-grade benchmark by CRR Research. Let's look at a specific run.
ProblemHere, a customer is requesting a flight cancellation. The agent processes the request and is about to approve the cancellation, but say this blocks it. By viewing the trace, we see the reason is that the customers on a basically economy ticket and the airlines policy defined an internal files does not allow cancellations for that fare class.
TractionSayless made the decision to block by validating the proposed action against company policy constraints and by checking for evidence from the actual booking. After blocking, sayless returns structured feedback to the agent regarding why the action was blocked and on how to best respond to the customer.
The agent course cracks and handles the request properly by escalating to a human agent. Across task-grade benchmarks for similar tasks, we've seen the 26% increase in correctness and that doubling in consistency compared to the same based on model without sayless.
And because sayless blocks bad actions early, it actually reduces cost per conversation by up to 60%. The same runtime-graderals also power RTBals. Through this tab, you can generate thousands of adversarial and realistic inputs for your agent and run them through the same validation checks so that you can improve your agent at the total-collable during development.
CTAIf you want to improve the correctness of your agents, please reach out. We'd love to give you a demo on board to you in just a few minutes. Thank you so much.
HookSequence Markets (YC W26) is building the execution layer bridging TradFi and DeFi across fragmented digital-asset markets.
How It WorksDigital-asset liquidity is fragmented across centralized and decentralized venues, each with different depth, fees, and latency. That fragmentation creates execution inconsistency and forces funds and active trading teams into manual, multi-system workflows.
Sequence gives teams one execution stack for routing, risk controls, and execution analytics across venues, built for execution quality and execution certainty under real volatility. They are building toward one-account, best-path execution across CEX, DEX, and tokenized rails over time, with API, SDK/MCP, and terminal workflows in one system.
Congrats on the launch, Peter Bai and Muhammad Awan!
https://lnkd.in/eyQdAX7g
HookThere are hundreds of places to buy and sell crypto. Each one has its own prices. Nothing talks to anything else.
ContextWhen an exchange goes down, people lose billions. Big players spread their money across dozens of platforms and lose a percent or more every time they move it. That's the market today.
Banks and blockchain. Different worlds same mess. We build a thing that connects them.
Secret markets. You place one order and we send it everywhere that matters at once. Every major exchange, every on-chain venue, and get you the best price.
12 microseconds from data to decision. 5 milliseconds there and back. One account, one outcome, every time.
How It WorksOne market. Markets weren't built to status broken. Where the connection.
Soon, everything trades like this. Real estate, loans, commodities, trillions moving on-chain. We're building the layer that runs it.
Every exchange, every venue, one execution engine. Cheaper than old school wire transfers. Open enough for programs and agents to trade on their own.
SolutionThe next chapter of market starts here. Sequence markets.
HookBreaking the best language learning apps. Do immersion. S.
Do immersion shows you short form videos in your target language that are exactly matched to your level and then you can just do a scroll like normal. If you don't know what a word means, you can click on the subtitles and get a definition and you can always scan the screen to figure out what those words mean to.
TractionYou can then swipe left to ask an AI tutor about those words and you can swipe right to see flashcards for any words that you've saved from videos. This way you can learn languages by immersion which is how people actually learn languages and you can do all doom scrolling which you really enjoy and also waste three hours a day on anyway.
CloseAlso it's my up and we're launching now so I'd appreciate if you can download the immersion on the app store and google play store. Plus we definitely have the cutest mascot.
HookZeroSettle (YC W26) is building a drop-in direct billing SDK for in-app purchases. It takes 15 minutes to integrate and instantly unlocks zero App Store fees, higher retention, and instant payouts for developers.
SolutionZeroSettle was born out of the famous Epic Games v. Apple lawsuit in May 2025 that required Apple to allow developers to offer direct billing. With $150B+ in annual in-app purchases, many mobile leaders are exploring ways to take advantage of it.
Current direct billing solutions often hurt conversion because they force users through web checkout during signup. ZeroSettle takes a different approach as it switches existing subscribers to direct billing with proactive offers. This avoids conversion risk and instead leads to massive margin improvements and improved retention.
TractionZeroSettle can also act as your MoR, handling the back office complexity Apple manages today, like chargebacks, fraud, and customer support. Or, you can connect your existing Stripe account.
The team is two former Apple engineers who worked on the core iOS and macOS frameworks used on billions of devices.
Congrats on the launch, Gabe Roeloffs and Ryan Elliott!
🚀 https://lnkd.in/g8uQ6n-s
HookZero-set-all is a mobile SDK that switches users from App Store billing to direct billing. It takes 15 minutes to integrate and developers immediately start enjoying zero App Store fees and higher retention.
ContextWe do this by intelligently showing offers to encourage loyal users to switch to direct billing. All the customer has to do is Apple Pay and Abraiser sheet and everyone wins. The users get a more affordable plan and devs get higher margins.
TractionDirect billing also improves retention. For example, if the user says they are canceling because it's too expensive, devs can configure a save the sale offer to keep retention high. You manage this growth engine from the zero-set-all dashboard.
You can see how many of your users are on direct billing in real time. And you can configure incentive discounts for your products and just a few clicks. When running Switch and Save campaigns, you can choose what percentage of the user base to show the offers to.
You can also target the campaigns based on how long someone has been a subscriber. So that's zero-set-all. Stop settling for a 70% cut and take back control of your revenue.
HookSo there's this new person. HR said their title is Cross-functional, which usually means gone in two weeks. But this one's different.
ContextDay one, infrastructure, day two, payments, day three, reading old Docs, nobody's touched since 2021. Walk past their desk at 9pm, still going. Old Slack threads from before I even joined.
SolutionMost dedicated higher-ever, or they're a spy, still not ruling off spy. The new higher-found Dave's old routing script, rebuilt it, then they looked at how we do triage and automated the whole thing. Tools that made us twice as fast.
One week. I've been on call five years. Every kind of outage, I don't get impressed easily.
I was impressed. 3am. Five teams, nobody sees the full picture.
How It WorksNew higher shows up and says, the 238 deploy changed three routing rules. Told us exactly what to roll back. 90 seconds.
Six years, never seen one person understand the entire system. It's kind of incredible. It's like they have the whole company in their head.
HookThe biggest bottleneck to AI co-workers joining the workforce is communication. Today we're launching talking computers. A new AI lab to make AI communicate just like humans.
SolutionOur first talking computer TC1 is a proactive self-improving general agent with what we call ambient intelligence. It can think in its subconscious, reach out to you for help, or even interrupt you whilst working on a problem together.
It is indistinguishable from a powerful remote worker inside of your business. We said refocus on communication. And in addition to agents being able to effectively communicate with humans, they should also be able to communicate with each other.
That's why we build facility. Through facility agents get a communication layer that they can use to talk to each other. It creates an intranet of all AI computers across your organization that are doing tasks.
Through facility TC1s get access to tools to send and receive messages, and also to a context layer that's constantly keeping everything in sync. And I mean everything. Every task, every decision, and every conversation that ever held the store.
The facility is where your employees and their TC1s can co-work together just the way a remote team does. They can get on calls, they can talk to each other, they can co-work on certain tasks, and they can get things done.
CloseThe company's building-off facility today are laying the foundation for an AI workforce that compounds over time. But we'll end with us today at talkingcomputers.ai.
HookUnifold (YC W26) is building universal deposit infrastructure for on-chain apps.
ProblemDeposits are still the biggest bottleneck in crypto. Even when users want to trade or participate, they drop off because funding flows force them through chain switching, bridge UIs, token approvals, and figuring out whether their USDC is on the “right” chain. For teams building consumer on-chain products, this kills conversion and makes multi-chain support painful to maintain.
How It WorksUnifold fixes this by giving developers a single API + SDK to accept on-chain deposits across any chain and token in less than 10 lines of code. They handle routing, gas abstraction, compliance checks, and settlement end-to-end, with native support for web and mobile.
They’re already helping leading prediction markets, perpetual exchanges, and on-chain consumer marketplaces deliver deposit experiences that users actually complete. Beyond established chains, they’re partnering with growing ecosystems such as Algorand, MegaETH, and Thru, with upcoming support for new networks like Tempo (from Stripe).
CTAIf you’re building a consumer-facing on-chain product and deposits are a bottleneck, reach out at @unifold_io or hello@unifold.io.
Congrats on the launch Timothy Chung, Hau Chu, and Quang H.!
Founder, Stealth <Data x Gen AI x B2B> • Product & Growth Advisor for AI Companies • Ex-Numeral (YC23), Wonderment
HookI'm going to show cross-layer labs detecting a hijack attack on a financial services platform. On the left, I have the cross-layer labs platform performing monitoring. In the middle, I have an adversary terminal, and on the right, I have a hypothetical financial services platform.
ContextOn the adversary runs this BGP attack command, a malicious BGP announcement is sent out to thousands of robbers across the internet, rerouting traffic to that financial services platform to the adversary. Cross-layer labs is monitoring the BGP layer and detects the suspicious BGP announcement.
However, the true plower of the cross-layer-level platform comes when the adversary requests a malicious certificate based on that BGP announcement. This certificate will let the adversary intercept and impersonate all communications to this financial services platform.
TractionThe adversary uses this to serve malicious JavaScript, which, when users go to load the financial services platform, will steal their cryptocurrency. The cross-layer labs platform is monitoring at multiple layers of the network stack, and also detects the suspicious new certificate, and most critically correlate that to the underlying BGP incident to give a high priority alert that also has root cause.
CloseIt does the spike in structuring a dependency graph so it can detect changes in different layers at the internet stack.
No screenshots available for this video.
2.72x
The Winning Formula
Free · The Launch Video Playbook
What to say. What to show. What to cut.
You've got the data. Now get the playbook — the opening lines the top 11 used word-for-word, the phrases that killed engagement, the shot order that held attention, and the pacing behind 1,378 median reactions. Everything the report didn't have room to show.
I help Y Combinator startups turn this data into a launch video that actually performs.
If you're launching soon and want a video built on what the data above says works — strong visual hook, tight editing, hybrid format, every second earning its place — let's talk.
Dataset: All 179 YC W26 launches posted to YC's LinkedIn account (Jan–Mar 2026). Of those, 173 had an analyzable video — 5 were image/text posts with no video and 1 had a malformed clip. All tier-level analysis (hook, editing, duration, style, etc.) is on the 173-video set; batch-wide totals in the hero (111K+ engagement, 503 median) are on the full 179. Engagement medians used where noted.
Engagement: Reactions + comments + reposts as of mid-April 2026.
Scoring: Every video was scored 1–5 on 6 dimensions: opening hook (first 3 seconds), camera/visual quality, editing & pacing, motion graphics/VFX, overall quality, and AI perception (how AI-generated the video looks). Initial scores were drafted by a multi-modal LLM pass and then manually reviewed frame-by-frame by the author — 98.9% of the ~960 dimension-level scores were overridden during human review, so what's reported is effectively a hand-scored dataset. Rubrics: hook = "does this earn the next 3 seconds of attention?"; editing = pacing and cut discipline, not polish; overall = the viewer's end-to-end impression.
Founder data: 382 individual LinkedIn profiles scraped via API across the full 200-company W26 batch; 175 of those companies also had a launch video in this analysis.
Limitations: Correlation, not causation. We can't isolate production quality from founder network, product novelty, or market timing. But the natural experiment (same YC account) controls for the biggest confounder: distribution advantage.
