Retrieval-Augmented Generation done right. Corpus design, chunking strategy, embedding model choice, hybrid retrieval (vector + BM25), reranking, query rewriting, citation, retrieval evaluation. Pairs with `evaluation-engineering` for measurement and with `agentic-architecture` for system integration. Most "the LLM made it up" failures are retrieval failures, not generation failures — get retrieval right and the generation problem shrinks dramatically. ML/AI Engineer owns this; Data Engineer co-designs the ingestion pipeline (per `data-pipeline`). Use whenever a feature needs the LLM to ground in specific documents, when designing the RAG corpus, when investigating low gr…
