Joe Maddalone introduces TypeSafe's Jev model, explaining its typed decision primitives (choice, score, boolean). He demonstrates a TypeScript script processing video transcripts for routing and scoring before calling traditional LLMs.
Original by Joe MaddaloneGetting startedIntermediate7 min 48 secPublished
Jev evaluates text input without generating freeform text, returning structured typed decisions via choice, score, and boolean primitives.
Running transcript analysis across multiple long files completed in seconds and cost under a penny according to the creator's test dashboard.
Jev can serve as a fast front-end classifier to filter, categorize, or score incoming data before routing to heavier LLMs.
Worth knowing
Subjective questions like evaluating whether an opening hook captures attention returned zero confidence, showing Jev may fail on ambiguous qualitative judgements.