A language model only knows what it saw in training. Retrieval-augmented generation (RAG) fixes that by finding relevant documents at question time and handing them to the model as context, which grounds its answers in your own data and cuts down on hallucinations. This module covers embeddings and vector search, how vector databases store and query them, and how to put the pieces together.
You'll build the core RAG pipeline of indexing, retrieval and generation, then look at techniques that improve it, such as query translation and routing, hybrid search, debugging, and observability. The final lesson focuses on taking RAG systems to production: scaling, cost, security, and advanced architectures.
These courses are long, so treat them as a reference: watch the parts you need, then come back for the rest. When you're done, continue on to AI Agents & Tool Use.
Vector embeddings and vector search explained, then three projects with Atlas Vector Search: semantic movie search, a question answering app using the RAG architecture, and a ChatGPT clone that answers from documentation.
The full RAG pipeline step by step: indexing, retrieval and generation, query translation (multi-query, RAG fusion, decomposition, step back, HyDE), routing, query construction, advanced indexing, and CRAG and adaptive RAG.
From prototype to production: document processing, embedding dimensions, Chroma, similarity search, hybrid search, debugging, LangSmith observability, scaling and costs, Supabase and PGVector, security, then agentic, graph and multimodal RAG.