RAG Systems

Retrieval-augmented generation is a loop: parse, clean, chunk, embed, store — then retrieve, assemble a prompt, answer with citations. This course builds that loop and then attacks it. You will measure recall@k rather than guessing, budget a context window so the important chunk is not the one that got cut, force citations so an answer can be checked, and work through the failure modes that make RAG systems quietly wrong: chunk boundaries, stale context, lost-in-the-middle, and conflicting sources. It ends where every good system ends — refusing to answer when the retrieved evidence does not support one.