MLOps
MLOps, short for machine learning operations, is a set of practices for building, deploying, monitoring, and maintaining machine learning models reliably in production. It brings data scientists, developers, and operations teams together to manage the full model lifecycle, from development and testing through deployment and ongoing monitoring.
MLOps applies ideas from DevOps to machine learning: version control for data and models, automated training and testing pipelines, repeatable deployment, and monitoring once a model is live. Because a model's accuracy can drift as real-world data changes, MLOps also covers detecting that drift and retraining.
For example, a vision AI model inspecting products may lose accuracy when a new packaging design is introduced. An MLOps pipeline detects the drop, gathers new labeled images, retrains the model, tests it, and deploys the update to the edge devices on the line.
MLOps differs from DevOps because it manages data and models, not only code. A software release behaves the same until it is changed, while a model can degrade on its own as the data it sees shifts.