Vector storage

Vector storage interviews appear when semantic search or RAG is in scope. Topics include embeddings, similarity metrics, ANN indexes, retrieval grounding, pgvector, and managed vector databases—often optional for general FullStack roles but useful for architecture discussions.

Subtopics: embeddings, similarity search, ANN indexes, RAG basics, pgvector, Pinecone / Weaviate / Qdrant.

On interviews: explain embedding pipeline, recall versus latency trade-off, metadata filtering with vectors, and when pgvector beats a dedicated store.

Common pitfalls: vectors without evaluation; ignoring chunking and citation; ANN parameters tuned without measuring recall.

The trade-off is balancing retrieval quality and latency against ops complexity and data governance for embeddings.

Checklist:

• Define embedding model, distance metric, and refresh policy.
• Filter by metadata before or with vector search.
• Measure recall/latency for ANN parameters.
• Choose pgvector vs dedicated store by ops and scale.