MultiPDF Chatbot is a fully local retrieval-augmented generation (RAG) system that transforms static PDF documents into an interactive knowledge base. Users upload one or more PDFs through a Streamlit interface, and the system indexes their content into a FAISS vector store — enabling real-time, context-aware question answering with conversational memory across multiple turns.
This system uses:
- Sentence Transformers (all-MiniLM-L6-v2) for embeddings
- Hugging Face LLM (Llama-3.1-8B-Instruct) for contextual and intelligent answers
- LangChain for retrieval orchestration and FAISS for efficient vector search
- Streamlit for a fast and interactive user experience It performs semantic retrieval using FAISS vector storage and maintains conversational memory for smooth multi-turn dialogue.
⚙️ Setup
- Create a virtual environment
- Install dependencies
- Create a .env file
- Run the Streamlit app
🧩 How It Works
- PDF Upload & Parsing: Extracts text from each uploaded PDF using PyPDF2.
- Text Chunking: Splits text into overlapping chunks (default: 1000 characters, 200 overlap) for contextual retrieval.
- Embedding Generation: Encodes text chunks into vector embeddings using the free model sentence-transformers/all-MiniLM-L6-v2.
- Vector Store Creation (FAISS): All embeddings are stored in a FAISS index for efficient semantic similarity search.
- Conversational Chain: When a user asks a question, the app retrieves the most relevant chunks, sends them with the query to the LLM (meta-llama/Llama-3.1-8B-Instruct via Hugging Face), and returns an answer.
- Memory & Context: The ConversationBufferMemory keeps prior chat history for coherent follow-up questions.