LLM Evaluation & Quality Gates

How to assess LLM output quality in production: golden test sets, human evals, model-based grading (LLM-as-judge), latency/cost trade-offs, and regression prevention. Critical for any role building AI-native products at scale — emphasized in Microsoft Copilot and Amazon GenAI platform roles.