SafeClaim

Inspiration

We got the idea for SafeClaim after seeing how stressful and confusing insurance claims can be. A few friends had car accidents and didn’t know what to say or do. Some accepted low offers, others said the wrong thing on record, or forgot to collect the right documents. Most people don’t realize how much those small mistakes cost them later.

We wanted to build something that could help right away, a simple and reliable tool that guides you through everything after an accident, step by step, so you don’t get taken advantage of and can get the settlement you actually deserve.

What it does

SafeClaim is your personal AI advocate after a car accident. It guides you step by step from what to do at the scene to getting a fair settlement. It helps file claims through a simple chat, reads police and insurance documents, estimates repair and payout amounts, and answers questions like “Is this settlement fair?” or “Do I need a lawyer?” SafeClaim makes the entire claims process fast, clear, and stress-free.

How we built it

We built SafeClaim with Next.js, TypeScript, and Tailwind for a clean, fast frontend. The backend runs on FastAPI with Supabase (PostgreSQL) for secure data handling. Gemini 2.0 Flash powers the AI that reads documents and chats naturally. Everything is containerized and deployed across Vercel and Fly.io for speed and reliability. Each AI agent runs independently, making the system modular and easy to scale.

Challenges we ran into

Getting multiple AI parts to talk to each other smoothly took a lot of trial and error. Parsing real police reports into clean data was tougher than expected, and making sure the AI gave safe, non-incriminating answers was a constant balance. We also had to deal with lag in the chat system while keeping the responses natural. Deploying both the frontend and backend within the hackathon time limit was its own challenge.

Accomplishments that we're proud of

We got a working end-to-end claim flow running in under 48 hours from accident input to claim generation. We built and deployed a live demo with chat, document reading. Most of it came together through nonstop late-night coding, quick fixes, and teamwork to get everything running smoothly.