MLOps
The set of practices for taking a machine learning model from a notebook to a reliably running production system — training pipelines, serving infrastructure, monitoring, and retraining, treated with the same rigor as regular software engineering (CI/CD, testing, observability).
MLOps borrows its name and much of its philosophy from DevOps: the goal is to make model training, deployment, and monitoring repeatable and automatable rather than a one-off notebook exercise.
In practice this usually covers: versioned training pipelines, automated evaluation gates before a model is promoted, a serving layer with monitoring for data/prediction drift, and a retraining loop that kicks in when performance degrades. It’s distinct from MLaaS (ML-as-a-Service), where a team consumes a hosted foundation model via an API instead of running its own training/serving stack.
Related posts
- MLaaS vs MLOps — 2023-02-01
- Functional Frameworks in Machine Learning — 2022-12-02