Training and deployment pipelines
Train from a config, evaluate against release gates and the production model, deploy in batch or behind an endpoint.
AI & ML
MLOps Templates
Pipelines to deploy, monitor, and retrain machine learning models.
Pipeline templates for training, validating, deploying, monitoring, and retraining machine learning models. Drift monitoring, model versioning, and rollback are included from the start. We use them for new models and to move existing models from notebooks into production.
Download the PDFWhat you get
Models that deploy routinely, alert you when accuracy drops, and are documented.
Free download · PDF · 10 pages
Get the MLOps Templates overview
The problem it solves, what is included, how it works, the technical components, and how we adapt it with you.
Your download has started.
We also emailed the link to . It stays valid for 7 days.
Download didn't start? Get the PDF
Want to see how it would fit your data? Talk to us.
The problem
Models stall between the notebook and production.
A model performs well in testing, then nobody can say which version is live, whether the data has shifted, or how to roll back. Retraining happens when someone remembers.
What this accelerator does
The MLOps Templates give every model the same release path: gated training, approved promotion, monitoring, and rollback from the start.
What's included
One release path for every model.
Four parts, each adapted to your data, platforms, and controls. What we adapt for you is yours to keep.
Feature pipeline and store patterns
Features defined once, point-in-time correct training sets, and a check for data leakage.
Monitoring
Drift, missing values, new categories, data freshness, accuracy, and cost, with alerts that can start retraining.
Evaluation Card and documentation
An Evaluation Card generated with every run and version, ready for the AI inventory.
How it works
The release path.
Train
Fits the pipeline and writes a run folder: model, metrics by group, data profile, config, and code version.
Gate
Every threshold met, gaps between groups within limits, and no meaningful drop against the production model.
Register
The run becomes the next version as a candidate, with its gate result attached.
Promote
Production needs a passing gate and an approver who is neither the trainer nor the registrant.
Roll back
One command restores the last archived version. Every change is logged.
Monitor
Scheduled checks compare live data with training data and alert when a limit is crossed.
Technical detail
Under the hood.
Vendor-neutral Python and configuration, Azure first, with tests included from the start.
- mlopskit package
- Training, gates, registry, scoring, monitoring, and cards on scikit-learn
- Model registry
- Stages, separate approver, history, and rollback; swaps for MLflow or Azure ML
- Drift monitoring
- Population stability index against the training profile
- Feature definitions
- Windowed features relative to an as-of time, with a leakage check
- Azure ML templates
- Command job, private online endpoint with Entra ID, blue/green releases
- CI/CD
- Train and gate on pull requests, approval for production, weekly drift check
Proof in the package
Worked example: customer churn, and the month after a price change.
A churn model is trained, gated, and promoted by a separate approver. Then the data shifts under it.
What happens
- The trainer tries to approve their own model and is refused
- Customer spend goes up
- Support tickets double
- A new plan code appears that the model has never seen
Result
Monitoring flags all three data changes and exits with an alert, so a scheduled job can open a ticket or start retraining.
How we run it with you
Adapted in the first cycles, handed over at the end.
- 01
Agree thresholds
Target, features, and release thresholds set with the Decision Owner.
- 02
Choose groups
The groups to compare, chosen with your governance lead, especially where decisions affect people.
- 03
Wire to your platform
Azure ML, Databricks, or MLflow registries and endpoints, with blue/green releases.
- 04
Monitor in production
Monitoring pointed at live scoring inputs, and at outcomes as they arrive.
You keep the pipelines, the registry history, the monitoring jobs, and an Evaluation Card for every model version.
Where it fits
Platforms, related accelerators, and limits.
We state the limits up front, and we recommend tools based on fit. We do not resell platforms.
Works with
- Azure Machine Learning
- Databricks Feature Engineering and MLflow
- GitHub Actions
Pairs with
- Trains and scores on the Data Platform Foundation
- Evaluation Cards feed the AI Governance Starter
- Uses certified metrics from the Metrics Layer
Assumptions and limits
- Training templates cover binary classification; the gate, registry, and monitoring carry over
- The folder registry suits one team; shared use moves to a platform registry
- Drift flags change, not harm, so alerts are reviewed with the owner
Take the overview with you.
Get the 10-page PDF to share with your team, or tell us the decision you want to improve and we will tell you whether the MLOps Templates fits.