SCOREAI infra
Jev Evals
AiIndigo@IndigoYogiArt
Jev evals replace expensive LLM judges with typed Jev decisions, enabling faster and cheaper evaluations within agent loops.
Pattern✦ Agent trace evaluation with typed judges
llm evals are too expensive. the judge model alone costs more than the thing you're evaluating.
jevals replaces LLM judges with typed Jev decisions. 8 evals, one request, $0.00006, milliseconds not seconds.
runs on every trace, inside the agent loop. open source.
github.com/openlayer-ai/jevals
