• Design model training and inference workflows with clear versioning, lineage, and promotion criteria where models are part of the system. • Define service responsibilities, interfaces, and data contracts that evolve safely. • Specify behavior under retries, timeouts, partial failures, and dependency degradation. • Choose consistency and durability guarantees that match risk, latency targets, and operational realities. • Design the request path for predictable tail latency and controlled resource usage. • Build and operate high-performance services and APIs that keep authentication reliable, secure, and fast at scale. • Implement distributed services that are safe under concurrency and robust to duplicate and out-of-order events. • Build real-time scoring and decision services with clear input/output contracts and bounded execution time. • Build distributed training pipelines that scale, are reproducible, and produce auditable artifacts. • Build pipelines that move data and model artifacts through validation, promotion, and release. • Define automated quality gates for service changes and releases. • Add checks for data quality, schema/contract adherence, and training-serving consistency where appropriate. • Define acceptance criteria tied to measurable outcomes and production behavior. • Coordinate multi-service releases with clear cutover and recovery plans. • Use production signals to validate rollouts and trigger rollback when risk is high. • Own after-hours production releases, including rollout validation, monitoring, and rollback execution. • Instrument the full path with metrics, logs, and traces that enable fast detection and diagnosis. • Implement alerting that reflects user impact, not just component health. • Lead incident response for your services, restore service quickly, and communicate clearly during events. • Drive reliability work through SLIs, SLOs, and error budgets, and make tradeoffs explicit. • Improve performance and cost through profiling, load testing, and capacity planning.