VORCL
← All courses
Intermediate · 7 lessons

Production RAG: retrieval systems that do not hallucinate

An advanced RAG course covering structure-aware chunking, hybrid search and reranking, query rewriting and multi-hop retrieval, agentic RAG, grounded answers with citations, evaluation, capacity planning, caching, SLOs, and safe rollouts.

Start →
  1. 01Chunking strategies: why naive splitting failsCompare fixed, recursive, structural, and semantic chunking and measure how boundaries, overlap, and document context affect recall.50 min
  2. 02Embeddings, hybrid search, and rerankersCombine dense and lexical retrieval with RRF and cross-encoder reranking, then tune the two-stage funnel against a golden dataset.55 min
  3. 03Query rewriting and multi-hop retrievalRewrite ambiguous questions, decompose multi-hop requests, and route queries without multiplying latency and cost without bounds.50 min
  4. 04Agentic RAG: when an agent should control retrievalGive an agent control over retrieval only where iterative planning adds value, with explicit budgets, stop conditions, and traceable state.50 min
  5. 05Grounding, citations, and hallucination controlRequire evidence-backed answers, verify citations mechanically, and measure faithfulness instead of trusting confident prose.50 min
  6. 06Evaluating retrieval quality: recall@k and faithfulnessSeparate retrieval from generation evaluation using recall@k, MRR, nDCG, context precision, and regression gates in CI.55 min
  7. 07Capstone: production RAG with an SLO, budget, and rollbackChoose a vector store, model capacity and cost, design tenant-safe caches and degradation, then prove canary and rollback in a game day.50 min

Sources

Production RAG: retrieval systems that do not hallucinate · VORCL