RAG Basic: Build an End-to-End RAG System

with Shirin and Shantanu

Build a complete Retrieval Augmented Generation system from scratch on a fully local, open-source stack — FastAPI, Docling, Sentence-Transformers, OpenSearch HNSW, vLLM, and Gradio, orchestrated as four Docker services. Your documents never leave your machine: no cloud APIs, no per-token billing, no vendor lock-in. 18 lessons take you from the first docker compose up through PDF ingestion, chunking, embeddings, vector search, and GPU-accelerated answer generation to a working web UI with an end-to-end test suite. Running the companion stack requires an NVIDIA GPU with roughly 16 GB of VRAM.