DeepEval adds JevEval: Jev as the judge for custom LLM evals, with no generative model in the loop
Confident AI’s DeepEval now ships JevEval, a custom metric judged by Jev. You write Noul, Score and Choice questions about a test case or an agent trace; Jev answers all of them in one request with probabilities; DeepEval takes a weighted mean. strict_mode returns 1 only if every check passes. The reason text is built in code from Jev’s answers, so it is deterministic. Built-in metrics also gained eval modes: llm, hybrid (Jev decides) and system_one (Jev only). It needs a TypeSafe key.
Surfaced by @Sumanth_077 on 13 Oct. This is an integration for Jev, not a new model. See Jev.