02 — Capability · BC-770.20
Machine Learning Engineering
Build, train, deploy and watch predictive models as production software, so a model that worked in a notebook keeps working in the business a year later.
- ML Engineering
- MLOps
- Predictive Modeling
In scope
- Training data preparation and feature engineering
- Model training with evaluation and selection
- Model deployment and serving
- Model monitoring and retraining
Out of scope
- Exploratory analysis and analytical modeling for one-off decisions (see BC-750)
- Infrastructure the models run on (see BC-730)
Realized by · 0
- No product in the catalog yet.
Used in · 0
- Not yet placed on a value stream.
Build it · 1
Decomposes into · 4
- BC-770.20.10Training Data PreparationAssemble, label and engineer the datasets models learn from, with their lineage and consent status known.
- BC-770.20.20Model Training & EvaluationTrain candidate models, compare them against honest holdout data and business metrics, and choose the one that earns deployment.
- BC-770.20.30Model Deployment & ServingPackage approved models and serve predictions through interfaces applications can rely on, with versioning and rollback.
- BC-770.20.40Model Monitoring & RetrainingWatch prediction quality and input drift in production and retrain or retire models before their advice quietly goes stale.