Whenever you're ready, walk me through how you'd approach this.
What metrics do you prioritize when performing LLM evaluation for a production-grade application?
Explain how you would select metrics for the application’s task, users, and risk profile. Discuss offline evaluation, human or expert review, automated judges, production monitoring, and how you would handle trade-offs between quality, safety, latency, cost, and reliability.