LangGraph RAG Pipeline (Self‑Healing, Azure‑Native)

This project implements a self‑healing Retrieval‑Augmented Generation (RAG) system using LangGraph, Azure OpenAI, Azure Blob Storage, and FAISS.

It supports:

• Automatic ingestion of PDFs from Azure Blob Storage
• Chunking + embedding using Azure OpenAI
• FAISS vector index creation
• Uploading vector store back to Blob Storage
• Query‑time retrieval and prompting
• A self‑healing mechanism that rebuilds FAISS only if missing
• An interactive CLI loop for back‑to‑back questions

---

🚀 Features

✔ Self‑Healing Vector Store

If faiss.index or metadata.json are missing, the system automatically:

1. Downloads PDFs
2. Extracts text
3. Splits into chunks
4. Embeds using Azure OpenAI
5. Builds FAISS index
6. Uploads index + metadata to Blob Storage

✔ Stateless Querying

Each question runs independently:

✔ Dynamic Prompting

Every question is injected into a fresh prompt with retrieved context.

✔ Back‑to‑Back Questions

The CLI loop lets you ask unlimited questions without rebuilding FAISS.

---

📁 Project Structure

---

🔧 Installation

1. Create a virtual environment

python -m venv venv
source venv/bin/activate   # macOS/Linux
venv\Scripts\activate      # Windows

2. Install dependencies

pip install -r requirements.txt

🔐 Configuration

cp config.env.example config.env

Fill all required config values.

• Azure OpenAI endpoint (AZURE_OPENAI_ENDPOINT)
• Azure OpenAI API key (AZURE_OPENAI_API_KEY)
• Embeddings deployment name (AZURE_OPENAI_EMBEDDINGS_DEPLOYMENT)
• Chat model deployment name (AZURE_OPENAI_CHAT_DEPLOYMENT)
• Azure Blob Storage connection string (AZURE_BLOB_CONNECTION)
• Containers for PDFs and vector store (AZURE_BLOB_CONTAINER)
• Containers for vector (FAISS) store (AZURE_BLOB_FAISS_CONTAINER)

💬 Interactive Question Loop

python main.py

You will see

RAG system ready. Type 'exit' to quit.

Ask a question: What are the top 2 policies name and premium ? Ask a question: What is the policy expire date ?

---

Ask a question: What are the top 2 policies name and premium ?

STATUS: Answered query using retrieved context. ANSWER: The top 2 policies are:

1. MARSH WALL KEY COVER - £109.00
2. MARSH WALL KEY COVER is not the second, actually it is COMBINED RTI AND FINANCE GAP - £594.00

So the top 2 policies by premium are:

1. COMBINED RTI AND FINANCE GAP - £594.00
2. MARSH WALL KEY COVER - £109.00

---