Retrieval-augmented generation as an engineering discipline: chunking strategies, embedding selection and versioning, vector stores (pgvector, dedicated engines) and ANN indexes, hybrid retrieval (BM25 + vector, reciprocal rank fusion), re-ranking, query rewriting, context assembly and token budgeting, grounding and citation, abstention, freshness and re-indexing, and retrieval evaluation (recall@k, precision, nDCG, faithfulness).
