ML in Production

ML in production means taking a trained machine-learning model out of a notebook and running it live, where it makes real predictions on fresh data for real users — for example scoring every incoming transaction for fraud. The discipline of doing this reliably is called MLOps (machine-learning operations).

It matters because a model that scored well in an experiment is only useful if it keeps working in the real world, and that part is surprisingly hard. Production demands serving predictions fast enough, monitoring for accuracy decay as the world changes (called drift), retraining on new data, and ensuring the features fed to the live model are computed exactly as they were during training. Skipping this is why many promising models never deliver value: the gap between “works in the notebook” and “works in production” is where most data-science effort actually goes.