Course 07 of 10
Retrieval
Retrieval is how a model answers questions about documents it has never seen. This course covers the whole path: cutting documents into pieces, embedding them, finding the right ones, and answering only from what was found. Every lab runs against a real index — the corpus is embedded and stored in pgvector, and the similarity scores are the ones your query actually produced.
Start lesson 01Foundations
- 01Why retrievalWhat a model knows, what it cannot know, and which of those problems retrieval solves.12 min
- 02ChunkingHow you cut the documents decides what can ever be retrieved.13 min
- 03Embeddings and similarityWhat a vector is, what 'similar' means to a model, and the asymmetry most people miss.13 min
Measuring
Answering
- 06Grounding and abstentionAnswering only from what was retrieved, and knowing when not to answer at all.14 min
- 07Evaluating RAGTwo components, two failure modes, and why one score cannot separate them.13 min
- 08RAG in productionFreshness, drift, cost, and the cases where retrieval is the wrong tool.12 min