AI-powered document chat system using LLaMA 3 and RAG architecture. Upload documents and ask questions with fully local inference (no API required). Built with FastAPI, Streamlit, and FAISS.

✨ Key Features

• 📄 Upload PDF, DOCX, TXT, MD documents

• 💬 ChatGPT-style conversational interface

• 🧠 100% local LLM (no API required)

• ⚡ Fast semantic search using FAISS

• 🔍 Source tracking for answers

💼 What This Project Demonstrates

🧠 Applied AI / LLM Engineering

• Built a Retrieval-Augmented Generation (RAG) system using a local LLM

• Integrated LLaMA 3 via Ollama for offline inference

• Designed prompts to reduce hallucination

🔍 Information Retrieval & Vector Search

• Semantic search using FAISS

• Transformer embeddings for document understanding

• Optimized chunking and retrieval

⚙️ Backend Engineering

• FastAPI-based backend

• Document ingestion pipeline

• Error handling and reliability

🎨 Frontend & UX Design

• Streamlit chat interface

• Chat history and file upload

• Backend-safe UX

🏗️ System Design

• Upload → Parse → Chunk → Embed → Store → Retrieve → Generate

🧪 Debugging & Reliability Engineering

• Solved race conditions

• Handled local model constraints

• Built fault-tolerant flow

🔐 Production Awareness

• Clean structure

• GitHub-ready setup

• Environment-safe practices

⚠️ Limitations

• High RAM usage

• No persistence

• Single user

🔮 Future Improvements

• Multi-doc support

• Chat memory

• Docker

• Cloud fallback

⚙️ How It Works

1. Upload document

2. Chunk & embed

3. Store in FAISS

4. Retrieve relevant chunks

5. LLM generates answer