Adani AI Labs – PDF RAG Conversational Agent

Objective

This project implements a Retrieval-Augmented Generation (RAG) based conversational agent that answers user questions strictly grounded in a provided PDF document.
The system retrieves relevant context from the document and generates answers only from that context, with clear citations.

---

Features

• PDF ingestion and page-wise text extraction
• Text chunking with page and chunk-level metadata
• Vector-based retrieval using FAISS
• Conversational (multi-turn) Q&A
• Grounded answers with citations (page / chunk references)
• Explicit refusal when information is not present in the document
• Retrieval visibility for debugging (top-k chunks + scores)

---

Tech Stack

• Python 3.10+
• Gemini API (for answer generation)
• FAISS (vector index)
• LangChain utilities (chunking & retrieval)
• PyPDF (PDF text extraction)

---

Project Structure

├── main.py ├── ingest.py ├── rag.py ├── llm.py ├── requirements.txt ├── README.md └── .env

Setup Instructions

1. Install dependencies

pip install -r requirements.txt

2. Set environment variable

Create a .env file or set the variable directly.

Windows

 
 setx GEMINI_API_KEY your_api_key_here

Mac / Linux

export GEMINI_API_KEY=your_api_key_here

Run the Application:

python main.py <path_to_pdf>