@_avichawla shows pg-jev, a Postgres extension that calls Jev from SQL with jev(), jev_prob(), jev_choice(), and jev_score(), and no vector column.

Walkthrough of an open-source extension. The post says semantic questions usually leave SQL, and pg-jev exposes Jev through SQL functions: jev() in WHERE, jev_prob() for a probability, jev_choice() for a label, jev_score() for ordered levels. It batches rows, returns typed answers, and needs no vector column or embedding index. The recording uses Hacker News stories in Postgres: find software-engineering stories, classify topic, rank technical depth, then report requests, tokens, cost, and cache hits for the session. Repo: https://github.com/realZachi/pg-jev. Includes a video and quotes their own article. Verified via fxtwitter.