Explore the deployment and monitoring architecture
To keep the no-show model reliable as it evolves, Proseware's team follows a machine learning operations (MLOps) process that spans everything from initial setup to ongoing monitoring.
Note
This diagram is a simplified representation of an MLOps architecture. To explore a more detailed set of implementations, see the MLOps v2 solution accelerator.
The architecture includes:
- Setup: Create the Azure resources the solution needs.
- Model development (inner loop): Explore data, and train and evaluate candidate models.
- Continuous integration: Package and register the model.
- Model deployment (outer loop): Deploy the model to an endpoint.
- Continuous deployment: Test the deployment and safely promote it to production.
- Monitoring: Track the endpoint's operational health and the model's predictions over time.
Registering a model is the hinge point between the first three stages and the last three. After the data science team trains and registers a model, the rest of this module focuses on that second half of the cycle: taking a registered model through deployment, safe promotion, automation, and monitoring.
Decide what to automate
You could perform each of these later stages manually in Azure Machine Learning studio. However, because Proseware plans to retrain and redeploy the no-show model regularly, the team automates registration, deployment, and testing with GitHub Actions, triggering the workflow whenever a new model version is ready. Automating these repetitive steps frees the team to focus on reviewing model quality rather than repeating deployment commands by hand.
Note
Automation is central to MLOps, but it doesn't replace human judgment. It's a best practice to keep a person in the loop for decisions such as promoting a new model version to full production traffic or investigating a monitoring alert.
The next unit shows how a protected GitHub environment turns that review point into an enforceable deployment gate.