Vector Databases

A vector database stores meaning. Where a traditional database finds rows by exact matches, a vector database finds items by semantic similarity — the content closest in meaning to a query. It is the storage and retrieval engine behind RAG, semantic search, recommendations, and agent memory.

The one-sentence version

Embeddings turn text into coordinates in “meaning space”; a vector database is the index that finds the nearest neighbors in that space, fast, at scale.

In this section

Embeddings What embedding vectors are, how embedding models produce them, dimensions, and similarity metrics.

Similarity Search & Indexing Approximate nearest neighbor search — HNSW, IVF — and the recall-vs-speed trade-off.

Choosing a Vector DB pgvector vs. dedicated databases, hybrid search, metadata filtering — and when you don't need one at all.

What you’ll be able to do

Explain how semantic search works under the hood, reason about the accuracy/latency trade-offs of an index, and choose — and justify — a vector store for a real workload.

Prerequisites

LLM Engineering, specifically the section on embeddings.