AI.info
MLOps
Production as a control system — contracts, risk tiers, ownership, and what happens at three in the morning.
- MLOps as a Control System
- Production Contracts, Risk Tiers, and Ownership
- ML Workload Architecture: Batch, Online, Streaming, and Edge
- Reproducible Environments, Configuration, and Secrets
- Versioning the ML Asset Graph
- Lineage, Provenance, and Release Evidence
- Experiment Tracking and Decision Records
- Registries, Approval, and Release Candidates
- Pipeline Design and Orchestration
- Idempotency, Caching, Backfills, and Recovery
- Testing Machine Learning Systems
- Model Packaging and Runtime Contracts
- Batch Inference Systems
- Online Inference Services
- Streaming and Event-Driven Inference
- Edge and On-Device MLOps
- Continuous Integration for ML Assets
- Continuous Delivery and Progressive Release
- Continuous Training and Retraining Policy
- Observability for ML Systems
- Service-Level Objectives and Error Budgets for ML
- Production Data Quality and Training–Serving Skew
- Model Quality Monitoring, Drift, and Delayed Labels
- Feedback, Labeling, and Human Review Loops
- ML Incident Response, Rollback, and Disaster Recovery
- ML Security and Supply-Chain Integrity
- Privacy, Governance, and Audit Operations
- Production Explainability, Documentation, and Decision Evidence
- Capacity, Cost, Performance, and Sustainable ML Operations
- ML Platform Engineering, Self-Service, and Multi-Tenancy
- Operating Foundation and Generative AI Systems
- Model Retirement, Decommissioning, and Evidence Retention
- MLOps Capstone: Design and Defend a Production System