How to monitor LLM drift, latency, and response quality in production. Best practices for maintaining healthy AI services.