The practices and tools for moving machine learning models into production and keeping them monitored and current once they are live.
MLOps (Machine Learning Operations) applies the habits of DevOps to machine learning, so the data engineers who prepare a model and the engineers who run it in production share one process. It focuses on standardizing the continuous delivery of AI models into production environments.
Building a model in a lab is comparatively easy; serving it securely and reliably to a large user base is much harder. MLOps frameworks automate the testing, deployment, and monitoring of models.
Without mature MLOps, technical drag builds up as data science teams struggle to move experimental models into the live enterprise architecture.
Mature MLOps also covers model versioning, continuous evaluation, data drift detection, and automated retraining, since models degrade in production as real-world data distributions shift over time.
See also Machine Learning, Technical Drag, Enterprise Architecture, Data Drift, Large Language Model and Fine-Tuning.