Implementing vector DB architecture isn't complex (so I'm bear on commercial vector dbs as a net value add especially since they all wrap some oss ANN solution)

Implementing vector DB architecture isn't complex (so I'm bear on commercial vector dbs as a net value add especially since they all wrap some oss ANN solution) You pair a vector db with a metadata store (can be anything, but ideally you want low gravity with the vdb and disk for fast retrieval... i.e. leveldb, sqlite equivalents... or hell a traditional db - and the author is right traditional dbs dont need to work too hard to create a ANN extension) In general, the greater challenge is the data engineering and overhead of managing retrieval stores (in terms of the data-integration/model pipelines) ... so I'm bull on the solutions addressing opportunities here.