Insurance Query Engine – Project Overview

An LLM-powered intelligent query–retrieval system designed to answer user questions from large, unstructured insurance documents such as policies, contracts, endorsements, and emails.

Purpose Enable fast, accurate, and contextual responses to insurance-related queries without manual document searching.

Core Functionality

• Document ingestion (PDF/Word/Text)
• Text chunking and semantic embedding
• Vector database creation for similarity search
• Retrieval-Augmented Generation (RAG) using LLMs
• Context-aware, explainable answers

Architecture

• Backend: FastAPI / Flask
• LLM: Gemini API
• Vector DB: FAISS / Chroma
• Database: MongoDB
• Frontend: React (Vite)

Key Features

• Accurate insurance policy Q&A
• Multi-document querying
• Reduced hallucinations via grounded retrieval
• Modular and scalable design
• API-driven backend with web-based UI

---

Setup Instructions

Backend

1. Run:

   pip install -r requirements.txt
   

2. Download required spaCy language model.

3. Run the backend:

   python main.py
   

4. To verify vector database creation:

   python test.py
   

Frontend

1. Navigate to frontend directory.

2. Install dependencies:

   npm install
   

3. Start the development server:

   npm run dev
   

Environment Configuration (.env)

MONGO_URI="your mongo uri"
GEMINI_API_KEY="your gemini api key"

Outcome

• Faster insurance query resolution
• Improved policy understanding for users and agents
• Practical application of LLMs + RAG in real-world insurance workflows