📚 Multi-PDF RAG Assistant

A fully local, API-key-free Retrieval-Augmented Generation system that lets you upload PDFs and ask questions about them with source citations.

🏗️ Architecture

PDF Upload → Text Extraction → Chunking → Embedding → ChromaDB Vector Store
                                                              ↓
User Query → Embedding → Semantic Search → Context Retrieval → LLM → Answer + Citations

🧩 Tech Stack

Component	Technology
Embeddings	BGE-small-en-v1.5
Vector Store	ChromaDB
LLM	LaMini-Flan-T5-248M (248M params, runs on CPU)
Framework	LangChain + Gradio

📊 Features

• 🔒 100% Local & Free — No API keys needed
• 📄 Multi-PDF Support — Query across multiple documents
• 📌 Source Citations — Answers include file name + page number
• ⚡ Lightweight — Runs on CPU with a 248M parameter model

🚀 Run Locally

pip install -r requirements.txt
python app.py

Then open http://localhost:7860

🧪 Evaluation

Evaluated using the RAGAS framework with GPT-4o-mini as judge, measuring:

• Faithfulness
• Response Relevancy
• Context Precision