AI & LLMs
What is a vector database?
A vector database stores text as numeric embeddings and retrieves the passages closest in meaning to a query, rather than matching keywords.
How the retrieval works
An embedding model converts each chunk of text into a list of numbers positioned so that similar meanings sit near each other. A query is embedded the same way, and the database returns its nearest neighbours. This is what lets a RAG system answer a question phrased in words that appear nowhere in the source document.
Which one to use
pgvector inside an existing Postgres is the right default for most business systems - one less service to run, and it joins against the data you already have. Qdrant and Weaviate are worth the extra infrastructure at large scale or with demanding filtering needs. Pinecone trades cost for having no infrastructure at all. The choice matters far less than chunking quality.
Where quality actually comes from
Chunking strategy and the embedding model dominate results. Chunks that split mid-argument retrieve badly no matter which database holds them. Overlapping chunks that respect document structure, plus stored metadata for filtering, plus a reranking pass over the top candidates - that is where the accuracy is, and none of it is a database feature.
When you do not need one
Under a few hundred documents, keyword search plus a large context window often beats a vector pipeline and is far simpler to operate. Vector search earns its complexity when the corpus is large, the queries are genuinely semantic, or you need filtered retrieval across structured metadata.
We build automation systems for small teams. Free audit call to map your specific workflows - no pitch, just a plan.