MotherDuck prompt_jev()
Classify text columns with typed Jev decisions directly in SQL.
Added to Jevfast
How it uses Jev
A SQL query sends each customer conversation transcript to MotherDuck's `prompt_jev` with a question and a typed Choice containing `billing`, `technical`, `sales`, and `account`. The returned classification label and confidence can then be used in SQL filtering, joins, and aggregations.
Why it is interesting
It brings typed Jev classification directly into SQL, so a query can label an entire table of text and immediately use those labels in ordinary analytical operations. The creator reports a benchmark of 100,000 AGNews rows in 40 seconds at 2,484 rows per second, 89% accuracy, and $0.50 cost; that result is creator-reported and not independently reproduced.