RAG notebook deep dive

Technologies Used

• LangChain: Chains the document loaders, splitters, vector store, and LLM.
• Hugging Face (Transformers & Embeddings): google/flan-t5-large for generation; sentence-transformers/all-MiniLM-L6-v2 for embeddings.
• FAISS: Vector database for fast semantic search over chunks.
• PyMuPDF: Parses uploaded PDFs into text.
• PyTorch: Backend for the Hugging Face models.
• ipywidgets: Interactive UI inside the notebook.

Notebook Flow

1. Setup: Installs dependencies, detects GPU.
2. Data Ingestion: Uploads a PDF and splits text with RecursiveCharacterTextSplitter.
3. Vector Database: Embeds chunks and stores them in FAISS.
4. Model Initialization: Loads flan-t5-large and creates a text-generation pipeline.
5. Basic RAG: Retrieves context and answers single-turn questions.
6. Conversational Memory: Rewrites follow-ups into standalone questions to keep chat history coherent.
7. Final Interface: Chat UI showing responses plus history.