-
-
Drag and Drop Policy upload
-
AI Chat Answer with Policy Reference
Inspiration
Millions of people buy insurance but 90% never understand their own policy. 50-page PDFs are full of legal jargon, hidden exclusions & complex waiting periods. When it's time to claim, claims get rejected. We wanted to solve this - What if you could just upload your policy and ask "Is my father's diabetes covered?" in simple English?
What it does
PolicyPilot is your AI Insurance Assistant:
- 📄 Upload any Health/Life/Motor insurance PDF
- 💬 Ask anything in plain English, Hindi, or Kannada
- 🤖 AI reads the entire policy and answers instantly with page references
- ✅ Tells you Coverage, Exclusions, Waiting Period, Claim Process, Network Hospitals
Example: "Meri policy me maternity cover hai kya? Kitna waiting period hai?" -> Instant answer with exact clause.
How we built it
- Frontend: React + Tailwind, PDF preview & chat interface
- Backend: Python FastAPI
- AI Engine: RAG (Retrieval Augmented Generation) - We chunk PDFs, create embeddings with Vector DB, and use LLM (GPT-4) to answer only from YOUR policy, no hallucination.
- OCR: PyMuPDF for scanned PDFs
- Deployed on: Vercel + Render
Challenges we ran into
- Scanned insurance PDFs had no text layer - solved with OCR
- LLM was making up answers - solved by strict RAG prompt: "Answer only from context"
- Handling 100+ page policies quickly - implemented smart chunking
Accomplishments that we're proud of
- Working end-to-end chat with PDF in 48 hours for BIET MineArena 1.0
- Accuracy >90% on real policies from Star Health, LIC, ICICI Lombard
- Supports English + Hindi queries
What we learned
- How RAG actually works vs just calling ChatGPT API
- Importance of UX for non-technical users (farmers, parents)
- How to handle large documents in LLM context window
What's next for PolicyPilot
- Add Voice Assistant for rural users
- Claim Letter Generator - Auto-draft claim emails
- Compare 2 policies side-by-side
- WhatsApp Bot integration
