Newpg-jev
Filter, rank and classify PostgreSQL rows using natural-language conditions.
Added to Jevfast
NewHow it uses Jev
The PostgreSQL extension sends row batches in a shared state with one yes/no Noul question per row, then exposes judgments as SQL predicates, probabilities, Choice results, or Score values.
What Jev decides
Support ticket teamExample answers · not a recorded Jev response · Source ↗
Question 1 · r0
YOUR APP
INSTRUCTION
For this record: which team should handle this?
STATE
A support-ticket row is included in the rows batch; Jev receives the row fields and the query as shared state.
JEV · CHOICE
- billing
- technical
- sales
App workflow
- Write a SQL expression
Use `jev`, `jev_prob`, `jev_choice`, or `jev_score` with a row and plain-language condition or rubric.
- Read ahead and batch rows
Stream eligible table rows and group them into shared Jev requests with one question per row.
- Return typed row results
Map returned judgments to booleans, probabilities, labels, scores, or full JSON results.
- Compose with PostgreSQL
Filter, order, group, join, and limit using the resulting SQL values.
Why it is interesting
It makes semantic row filtering and classification composable with normal SQL operations, while batching, read-ahead, caching, and concurrency handle query execution.