Whenever you're ready, walk me through how you'd approach this.
How would you evaluate the performance of an LLM in a production environment? Describe a practical evaluation framework covering offline test sets, human and automated quality assessment, safety, latency, cost, reliability, user feedback, and data or behavior drift. Explain how you would establish baselines, choose metrics, monitor regressions, and use evaluation results to guide model or prompt changes.