Racoonn started as an internal tool. We kept running expensive, slow user studies — and still shipping products with obvious friction points nobody caught. So we built the thing we wanted.
K
Kevin Lin
Co-founder & CEO
Previously led product at NVIDIA's AI platform team, where he spent four years obsessing over developer experience and why highly technical products confuse smart people. Before that, Stanford CS.
🟢 ex-NVIDIA · Stanford CS
S
Sarah Park
Co-founder & CTO
Three years at Anthropic working on model evaluation and behavioral simulation. She wrote the original persona engine that powers Racoonn — the idea was "what if we ran evals on UX instead of model outputs."
🟣 ex-Anthropic · ML researcher
M
Marcus Yeh
Head of Design
Led design for Linear's onboarding and Vercel's first-time user experience. Has an unhealthy obsession with the exact moment users decide a product isn't for them — and what could have changed their mind.
🔵 ex-Linear · ex-Vercel
J
Jamie Torres
Growth & Partnerships
Previously at Stripe running PLG experiments and at Figma during their community-led growth phase. Joined Racoonn because she believes the next wave of B2B growth tools will be AI-native from the ground up.
🟩 ex-Stripe · ex-Figma
Why we built this
At NVIDIA, Kevin watched teams spend $40K on user research for a developer tool — only to get a 40-page PDF that said "users found the onboarding complex." At Anthropic, Sarah realized that behavioral simulation at scale was technically possible, just nobody had applied it to product UX.
The insight: user testing isn't broken because it's hard to do — it's broken because the incentives are wrong. Recruiting real users takes weeks. You test when you have budget, not when you have questions. And the people you test are never the people who actually left.
Racoonn runs 5,000 AI personas — each built from real demographic and behavioral data — through your product. They arrive with different goals, different patience levels, different devices. They judge. They drop off. You get a report with ranked issues and specific fixes.
We use it on our own landing page every week. The version you're looking at right now has been tested by 5,000 AI users. (We found three things to fix.)