Vector database
A vector database stores embeddings—fixed-length numeric vectors—and optimizes similarity search: find nearest neighbors by distance (cosine, Euclidean, dot product), often with metadata filters alongside.
Use it when queries look like “which items are most like this vector?” rather than “which keys equal X?” Typical workloads: RAG retrieval, recommendations, deduplication, semantic search.
Theory that matters here
Vector databases revolve around embeddings, distance metrics, approximate nearest-neighbor indexes, metadata filtering, index freshness, and recall/latency tradeoffs. HNSW and IVF are common index families in this space.