Glossary
MLOps (machine learning operations)
The set of practices for deploying, monitoring and maintaining machine learning models reliably in production, extending DevOps to ML systems.
MLOps is the set of practices and tooling for deploying, monitoring and maintaining machine learning models reliably in production, applying ideas from software DevOps to the specific challenges of ML systems: models depend on data as well as code, and their performance can degrade silently even when nothing was deployed.
A typical MLOps setup covers versioning both code and training data, automated pipelines for retraining and deployment, a feature store to keep training and live features consistent, and model monitoring to catch accuracy loss from data drift or concept drift after launch. It also addresses reproducibility, being able to recreate exactly which data and code produced a given model, and governance, tracking which model version is serving which decision.
MLOps matters because a model that performs well in an offline evaluation can still fail in production due to pipeline bugs, drifting inputs, or infrastructure that cannot handle live traffic; without it, models are difficult to update safely or audit after the fact. Organizations without mature MLOps often see models quietly go stale, or need weeks of manual effort to retrain and redeploy something that should be routine, which is why it has become a distinct discipline alongside data engineering and data science.
Last reviewed September 22, 2026