CHOOSEAgents & automation
Jev Decision Layer
Jersey | AI projects@vpsproof
Jev is a decision layer for LLM applications that uses three primitives: Choice, Score, and Noul, to avoid hallucinations and type errors by emitting no strings.
TakeJev not a chat model but a decision layer with three primitives: Choice (pick one of your options, returns the full probability distribution), Score (rate on an ordered scale), Noul (yes/no with a calibrated probability).
Pattern✦ Decision layer primitives
Jev not a chat model but a decision layer with three primitives:
Choice (pick one of your options, returns the full probability distribution), Score (rate on an ordered scale), Noul (yes/no with a calibrated probability).
It emits no strings at all, so hallucination and type errors are structurally impossible.
Value for builders: most LLM calls in a product aren't "write something", they're "which bucket does this belong to" — support classification, task routing, agent routing, refund-intent detection.
Browser Use's sample answers "which action" and "which element" in one round-trip, calling a text model only when typing is actually needed.
Mechanism and benchmarks included; the model is still in early access.
Related builds
