Course
RAG in Production
Get a model to answer from your own documents without making things up: chunking, embeddings, vector database, reranking, and answer evaluation.
Advanced 8 lessons
Lessons
- 1 Why RAG You will be able to identify the problems a RAG system actually solves, the ones it will never solve, and where the dominant cause of its failures lies.
- 2 Chunking documents You'll learn how to choose a chunking strategy suited to your documents, handle tables and code, and measure the effect of chunking on recall.
- 3 Embeddings and similarity You'll be able to choose an embedding model suited to your corpus, understand what cosine similarity measures, and avoid the pitfalls of multilingual and very short texts.
- 4 Vector database You'll learn how to choose between pgvector, an in-memory index, and a dedicated service, tune approximate indexing, and design metadata that works for filtering.
- 5 Hybrid Search You'll learn how to combine BM25 and vector search with rank fusion, add a reranking model, and measure the real gain from each stage.
- 6 Building the prompt You'll learn how to structure a RAG prompt that enforces citation, refuses to answer outside the context, and takes advantage of passage order in the context window.
- 7 Evaluating answers You'll learn to build a reference question set, measure retrieval recall and answer quality separately, and detect unsupported claims.
- 8 Going to production You will learn how to calculate cost per query, cache in the right places, reindex incrementally, delete a document cleanly, and monitor the system in operation.