Demystifying Insurance Policies with AI

PolicyPal is an intelligent web application designed to decode complex insurance documents using the power of artificial intelligence. Upload any policy PDF, ask your question in plain English, and receive a clear, fact-grounded answer — instantly, accurately, and transparently.

🔗 Live Demo: Try PolicyPal Now

Understanding insurance policies shouldn't require legal expertise. PolicyPal bridges the gap between dense insurance jargon and everyday clarity, powered by Retrieval-Augmented Generation (RAG) and best-in-class LLM APIs.

This repository contains the fully deployed version of PolicyPal — a refined result of technical agility and architectural decisions made during development.

Our original vision centered around a fully self-hosted AI pipeline powered by Google's open-source Gemma model. While we successfully built this system (available here), deployment limitations on available infrastructure prompted a strategic pivot.

To ensure a seamless user experience, we transitioned to cloud-based APIs for inference and embeddings, allowing us to showcase the application's full potential — without compromising its logic, responsiveness, or integrity.

💡 Our self-hosted version remains the technical foundation of this project. Explore it here:
🔗 Gemma Architecture (Self-Hosted)

This version follows a clean, modular, three-tier microservice architecture:

• Framework: React.js
• Role: Provides a responsive, elegant interface for file uploads and user queries
• Hosting: Vercel

• Framework: Node.js (Express.js)
• Role: Serves as a secure API gateway between the client and the AI service
• Hosting: Render

• Framework: FastAPI (Python)
• Role: Core document question-answering logic using RAG
• APIs Used:
  • Inference: Groq (Llama 3)
  • Embeddings: Cohere
• Hosting: Render

Set up the entire system on your machine in minutes.

• Node.js (v18 or later)
• Python (v3.10 or later)
• API Keys from:
  • Groq
  • Cohere

git clone https://github.com/muskan-khushi/PolicyPal-Deployed.git
cd PolicyPal-Deployed

Create an environment file for the AI service:

# /doc_qa_backend/.env
GROQ_API_KEY="your_groq_api_key"
COHERE_API_KEY="your_cohere_api_key"

cd doc_qa_backend
python -m venv venv
# On Windows:
venv\Scripts\activate
# On Mac/Linux:
source venv/bin/activate

pip install -r requirements.txt
uvicorn app.main:app --reload --port 8000

Service runs at: http://localhost:8000

cd ../server
npm install
npm start

Server runs at: http://localhost:5000

cd ../client
npm install
npm start

App opens at: http://localhost:1234

PolicyPal represents the intersection of practical AI engineering and real-world problem solving. What started as an ambitious self-hosted AI project evolved into a production-ready application that demonstrates both technical depth and deployment pragmatism. Key Achievements:

✅ End-to-end RAG implementation from document processing to response generation ✅ Production deployment across multiple cloud platforms ✅ Architectural flexibility - seamless transition from self-hosted to cloud APIs ✅ User-centric design - complex insurance logic translated into clear, actionable insights

This project showcases not just the ability to build sophisticated AI systems, but the engineering judgment to adapt and deploy them effectively in real-world constraints. PolicyPal makes insurance accessible, one query at a time.

Built with ❤️ and lots of ☕