panic.design

Drop any UI screenshot or URL. See exactly where 5 real user types freeze, rage-click, or leave - before your real users do.

Comment

Inspiration

Most UX feedback comes too late, costs too much, or tells you things you already knew. I kept seeing the same problem. Teams spending hours debating button colors while their onboarding flow was silently losing 60% of users on step two. Nobody catches it until real people start bouncing. I built panic.design to catch it before they do.

What it does

You upload a screenshot of your interface or paste a live URL. panic.design runs it through five distinct user personas representing real categories of user behavior that existing tools ignore. The Anxious User who second-guesses every click. The Distracted User who gave your page three seconds. The First-Timer with no context. The Impatient Mobile User on a slow connection. The Skeptic who assumes every interface is trying to trick them.

Each persona flags specific friction points, speaks their complaints in first person, and produces a sub-score. The app combines these into a single Panic Score. The score places your UI in one of five tiers: Panic-Proof, Steady, Work In Progress, Stress Fractures, or Crime Scene. Every complaint comes with a Fix Generator that gives you the actual rewrite, not advice.

How we built it

The AI analysis runs on Gemini Vision via Google AI Studio and its IDE. The URL input is powered by Nimble’s live web rendering API, which converts any live website into a screenshot automatically. The frontend is deployed on Vercel. The entire stack is free.

Challenges we ran into

Getting five personas to produce genuinely distinct voices and not collapse into the same generic UX feedback was the hardest part. Each persona needed its own reasoning pattern, not just a different tone. Prompting Gemini to maintain that separation consistently across wildly different UI types took significant iteration.

Accomplishments that we're proud of

The Before/After mode. Upload v1 and v2 of your interface and the app tells you which problems were solved, which stayed, and whether you introduced new ones. No existing tool does this at zero cost. The Panic Score animation and the Crime Scene verdict card also hit harder than expected in demos.

What we learned

Users do not fail at interfaces randomly. They fail in predictable patterns based on who they are. Building around that insight instead of generic heuristics made every output feel specific and honest rather than templated. Also Nimble’s rendering API is genuinely fast.

What's next for pain.design

Industry-specific persona packs. An e-commerce skeptic behaves differently from a SaaS skeptic. Mobile-only audit mode. A shareable report link so teams can send panic.design results directly to their designers without screenshots. And a freemium model at $19 per month for unlimited audits.

Built With

Share this project:

Updates