Insurance Policy Intelligence Engine

Upload an insurance policy (PDF or text) and ask questions about it in plain English, from a React dashboard on Cloudflare Pages backed by a Python Cloudflare Worker at the edge.

Live: empire-rag-dashboard.pages.dev · API health: /health

What it does

1. Client-side document parsing. PDFs are parsed in the browser with PDF.js, so the raw file never leaves the user's machine; only the extracted text is sent.
2. Edge ingestion. POST /api/upload stores the document text in the Worker.
3. Grounded answers. POST /api/chat splits each document into paragraphs, retrieves the ones that match the question's terms, and returns them with the source filename. Answers are always excerpts of the actual policy, never generated text, so the system can't hallucinate coverage that isn't in the document.

Architecture

React + TypeScript dashboard (Cloudflare Pages)
  │  PDF.js text extraction in the browser
  │
  ├── POST /api/upload  {filename, text}  ─┐
  ├── POST /api/chat    {query}           ─┤──► Python Cloudflare Worker (entry.py)
  └── GET  /health                        ─┘      paragraph split → keyword retrieval → cited excerpts

Path	Role
entry.py	Worker: routing, CORS, upload, retrieval, cited answers
wrangler.toml	Worker config (Python runtime; Workers AI and Vectorize bindings declared)
dashboard/	React + TypeScript + Vite + Tailwind UI: upload, PDF parsing, chat, source display

Current scope and roadmap

Retrieval is keyword-based over an in-memory store, so documents last only for the lifetime of a Worker instance. The Workers AI and Vectorize bindings are already declared in wrangler.toml for the next step:

• Chunk and embed documents with Workers AI (@cf/baai/bge-base-en-v1.5) into Vectorize
• Top-k semantic retrieval in place of keyword matching
• Llama 3 on Workers AI to synthesize an answer from retrieved chunks, with citations
• Persist documents in R2/KV instead of instance memory
• Compare plans across several uploaded policies

Run it

Frontend (talks to the deployed Worker):

cd dashboard
npm install
npm run dev          # http://localhost:5173

Backend:

npm install -g wrangler && wrangler login
npx wrangler deploy                                           # from the repo root
curl https://empire-rag-backend.soorya220104.workers.dev/health

Deploy the dashboard:

cd dashboard && npm run build
npx wrangler pages deploy dist --project-name empire-rag-dashboard

If the Worker URL changes, update BACKEND_URL in dashboard/src/App.tsx. More commands are in COMMANDS.md and DEPLOYMENT_GUIDE.md.

Tech stack

Cloudflare Workers (Python runtime) · Cloudflare Pages · React · TypeScript · Vite · Tailwind CSS · PDF.js

License

MIT — see LICENSE.