Most modern semantic search systems use vector similarity search. A nice property of semantic search is that most of the computation (embedding the chunks) can happen ahead of time, and the system only needs to generate the single query embedding at query time. I set up a cloud hosted Postgres database and installed the PGVector vector search extension, then uploaded all the data. In pure vector search with PGVector, this is easy using SQL’s familiar LIMIT and OFFSET syntax. More relevant search results.