Vector databases and retrieval
A vector database stores meaning rather than keywords, which is what lets a system find the right passage when the user's words do not match the document's words.
Where we use it
Choosing one matters less than teams think. Postgres with the pgvector extension handles a surprising number of real workloads and avoids adding another system to operate. Dedicated vector databases earn their place at high volume or with demanding filtering requirements.
What actually determines quality is what surrounds the database: chunking, the metadata attached to each chunk, query rewriting, hybrid keyword-plus-vector search, and reranking before results reach the model.
- Embedding strategy and model selection
- Chunking tuned per document type
- Metadata filtering and permission enforcement
- Hybrid search combining keyword and semantic
- Reranking for precision at the top of results
- Index refresh, versioning and cost monitoring
Choosing one, and leaving one
For most projects the honest answer is Postgres with pgvector: you already run it, you already back it up, and it handles hundreds of thousands of chunks without complaint. Dedicated vector databases earn their place at genuine scale, with heavy metadata filtering, or when you need features Postgres does not have.
Whichever it is, the retrieval layer sits behind an interface so the store can be swapped without rewriting the application. Providers in this market are young and their pricing changes; being able to move is worth more than a small performance edge.
What actually determines retrieval quality
Rarely the database. It is chunking strategy, whether metadata is captured at ingestion, whether you rerank, and whether you measure. A well-tuned pipeline on a mediocre store beats a careless one on the best store available.
Embeddings also drift: change the model and every existing vector becomes incomparable, so re-indexing has to be a routine operation rather than an emergency. That is a design decision made at the start.
Frequently asked questions
Which vector database should we use?
Start with pgvector if you already run Postgres — one less system to operate. Move to a dedicated store when scale, filtering complexity or latency requirements justify it, not before.
How much does it cost to run?
For most business knowledge bases, tens of dollars a month. Costs grow with embedding volume and query rate, both of which we model during scoping.
Tell us what you are building.
Send a short description of the problem and we will reply within one business day with an honest view of scope, cost and whether we are the right person for it.
Or email directly: contact@hire-ai-dev.com