Twelve curated topics every AI engineer needs: pgvector cosine/L2/inner-product distance, HNSW vs IVFFlat vs Flat exact indexes, Pinecone serverless/pod topology, Weaviate hybrid BM25+vector search, Chroma embeddings persistence, FAISS IVF-PQ scale patterns, metadata filtering pre- and post-search, hybrid lexical+semantic retrieval, chunking strategies (fixed, sentence, semantic, late), embedding model tradeoffs, ANN evaluation (recall@k, QPS), and production failure playbooks.