# MLOps
MLOps applies [[DevOps]] principles to machine-learning systems: version-controlled code, automated training pipelines, model registries, CI/CD for models, monitoring in production, and rollback strategies. The point is to make the path from "model trained in a notebook" to "model serving live traffic, reproducibly" boring — not heroic.
MLOps differs from DevOps in three ways: the artifact is not just code but **code + data + model**, performance degrades silently as data drifts, and experiments are first-class citizens with their own lifecycle.
## References
## Related
- [[MLflow]]
- [[Model Registry]]
- [[ML Reproducibility]]
- [[ML Deployment Patterns]]
- [[LLM Monitoring]]
- [[DevOps]]