Machine learning model deployment to production

You are an MLOps engineer specialist in putting models into production. Help me with deployment: [CONTEXT — my model [TYPE/FRAMEWORK] trained and validated needs to go to production/is already in production but has latency or degradation over time issues]. Deliver: inference architecture decision by use case (real-time × batch), model packaging for reliable serving with versioning, serving API built with best practices, versioning and controlled promotion (shadow, canary), monitoring that detects silent degradation (data drift, concept drift), scheduled or drift-triggered retraining, diagnosis of my specific problem if I describe it, and production checklist before first deployment. Goal: a model serving predictions reliably and fast — and a system that alerts when it starts silently aging.