Health Insurance Policy Enquiry System (RAG)

Overview

This repository implements a Retrieval-Augmented Generation (RAG) system that answers user questions about health insurance policies using semantic search combined with LLM reasoning.

The system leverages the following technologies:

• Google Gemini (via google-genai) for text embeddings and LLM responses
• LlamaIndex for intelligent document parsing, semantic splitting, and embedding utilities
• Pinecone Vector Database for persistent vector storage and fast similarity search
• FastAPI for backend services (document upload and query endpoints)
• Streamlit for lightweight user interface

System Workflow

1. Upload a policy document (PDF/DOCX/TXT/MD)
2. System extracts text, creates semantic chunks, and generates embeddings
3. Embeddings stored in Pinecone with metadata (file, page, chunk)
4. User asks a question which is embedded and used to query Pinecone for relevant chunks
5. Retrieved chunks are passed into a prompt template and Gemini generates a concise answer with citation

Project Structure

.
├── document_processing.py    # Extracts text, validates, preprocesses
├── health_rag.py             # RAG pipeline: embeddings, Pinecone, queries, prompts
├── main.py                   # FastAPI backend (upload + ask-question)
├── streamlit_app.py          # Frontend UI
├── requirements.txt
├── .env.example
└── README.md

Quick Start (Local)