Deploy and serve models on Databricks Model Serving and self-hosted inference such as vLLM, and manage the lifecycle in MLflow, including tracking, tracing, registry, release management, and CI/CD for models, prompts, and agents with shadow, canary, and A/B rollouts. Build evaluation and monitoring tied to business outcomes, including golden datasets, field-level precision, recall, and F1, retrieval metrics, groundedness, LLM-as-judge calibrated to human labels, release-gating regression suites, and production drift and agent-trace monitoring.