Playground for Jev and Laya System One models — classify text, route intents, score inputs, and test LLM guardrails with structured judgments.