Multi-Document AI Assistant

A powerful AI assistant capable of answering questions from multiple PDF manuals with source citations using LangChain, FAISS, and Google Gemini 2.0 Flash Lite. Give it a try here! (Might take a second to load.)

Note: Still a WIP, the retrieval component could frankly be a lot better. Also bear in mind a lower end LLM model is used due to cost constraints so please verify any answer you get from this.

Features

• Multi-PDF Processing: Upload and process multiple PDF documents
• Intelligent Text Chunking: Smart text segmentation for optimal context retrieval
• FAISS Vector Store: Fast and efficient similarity search
• Source Citations: Every answer includes references to source documents
• Streamlit Web Interface: User-friendly web application
• Google Gemini 2.0 Flash Lite Integration: Advanced language model for accurate responses

Project Structure

• app.py: Main Streamlit application
• pdf_processor.py: PDF processing and text extraction
• vector_store.py: FAISS vector store management
• qa_system.py: Question answering system with LangChain
• utils.py: Utility functions
• requirements.txt: Python dependencies

How It Works

1. Document Processing: PDFs are parsed and text is extracted
2. Text Chunking: Text is split into manageable chunks with overlap
3. Embedding Generation: Chunks are converted to vector embeddings
4. Vector Storage: Embeddings are stored in FAISS for fast retrieval
5. Question Answering: User questions are processed through the retrieval-augmented generation pipeline
6. Source Citation: Relevant source documents are cited in responses