Concept¶

Introduction to MLOps.

What is MLOps?¶

MLOps (Machine Learning Operations) is the set of practices that bridges the gap between machine learning development and production operations. It turns a machine learning experiment into a system that can be reproduced, deployed, monitored, and improved over time.¹

It sits at the intersection of machine learning, DevOps, and data engineering:

MLOps applies version control, automated testing, continuous integration, and continuous deployment to machine learning. The goal is a reproducible pipeline around the model that lets teams share experiments, deploy confidently, monitor performance, and retrain when production data drifts.

The scope of an ML system¶

A common mistake is to treat the model as the product. The model code and trained weights are only a small part of a production machine learning system. The surrounding infrastructure is far larger: data pipelines, training orchestration, experiment tracking, serving infrastructure, monitoring, and feedback loops.²

MLOps provides the tools and discipline to manage this infrastructure. The goal is to build a reliable system around the model rather than tune the model on its own.

What MLOps addresses¶

These challenges arise when an experiment must be shared, deployed, or maintained:

• Reproducibility. The same experiment must produce the same result for a different person, on a different machine, or months later. Version control and data versioning ensure the same result every time.
• Automation. Manual training, evaluation, and reporting steps do not scale across a team. A continuous integration pipeline automates these steps and validates every change.
• Deployment. A trained model is useful only when it can serve other systems. Packaging it into a container and deploying it to Kubernetes makes it available.
• Monitoring. Production data changes, and model performance degrades over time. Monitoring detects drift and signals when a model needs attention.
• Retraining. New data closes the loop. Labeling it with AI assistance and retraining the model improves the system over time.

MLOps turns experiments into systems that teams can reproduce, deploy, and improve.