ML Platform
This project assembles a small ML platform for a team that needs reproducible training, evaluation, scheduled or on-demand batch inference, and observable operations without a large platform team. The course develops the implementation from a local path toward Azure deployment.
What it contains
Training and evaluation jobs record their runs and artifacts through MLflow. A PostgreSQL results database tracks operational state across job types. Batch jobs and an online serving app use versioned model artifacts, while a dashboard exposes results and job launches. The repository includes containers, Terraform modules, deployment scripts, smoke tests, and CI configuration for Azure Container Apps.
The design keeps execution, model lifecycle, operational state, and serving as separate concerns. Managed identities provide machine authentication in the Azure path; the local course path can be studied without a cloud deployment.
Explore the project
The infrastructure code describes a deployment path, not a claim that a public service is currently running. Reproduction costs depend on the Azure resources selected for a run.