What is MLOps

Definition

MLOps is the set of practices that applies DevOps principles to machine learning, automating and standardising the lifecycle of building, deploying, monitoring, and retraining models so that ML systems are reliable, reproducible, and maintainable in production.

« Back to Glossary Index

Automates the path from model development to reliable production deployment
Makes ML reproducible through versioned data, code, and models
Enables continuous monitoring and retraining as data and performance change
Bridges data science and operations for maintainable ML systems

Real World Example

A retailer adopts MLOps so its demand-forecasting model is retrained automatically as new sales data arrives, validated against metrics in a pipeline, and deployed with rollback, replacing a fragile manual hand-off between teams.

FAQs

How does MLOps differ from DevOps?

MLOps extends DevOps with ML-specific concerns like data and model versioning, retraining, and monitoring for model drift.

What problems does MLOps solve?

It addresses the difficulty of deploying, monitoring, reproducing, and updating models reliably once they leave the notebook.

What does an MLOps pipeline include?

Data and feature management, training, validation, deployment, monitoring, and automated retraining of models.