Individual MLOps coursework · 2026
Azure ML lifecycle
A reproducible machine-learning workflow from cloud data preparation and training to versioned models, containerized inference, deployment manifests, and automated verification.
Lifecycle
More than model training
The coursework treated the model as one part of a larger system. Azure resources, data assets, environments, pipeline components, model versions, inference services, and deployment checks were represented explicitly.
Stage 1Azure data assetsFifteen classes prepared in parallel
Stage 2Data splitCombined, balanced 80/20 train-test split
Stage 3Cloud trainingVersioned Azure ML environment and compute cluster
↓ metrics + artifacts ↓
Stage 4Model registryVersioned model and associated metadata
ServingFastAPI inferencePredictions, health endpoint and PostgreSQL history
DeliveryCI/CD workflowBuild, deploy, health check and notification
Implementation
Concrete cloud and platform work
- Provisioned an Azure ML workspace and compute cluster and defined reusable command components and environments.
- Built a four-stage Azure ML pipeline: parallel preprocessing, splitting, CNN training, and model registration.
- Used MLflow-compatible tracking and versioned model artifacts rather than treating a local file as the deployment boundary.
- Built a FastAPI inference service and Streamlit Grad-CAM dashboard backed by PostgreSQL prediction history.
- Containerized the system with Docker Compose and authored Kubernetes deployments, services, persistent volumes, secrets, resource limits, and health probes.
- Automated setup, training, model download, local deployment, health verification, cleanup, and Discord status notifications with GitHub Actions.
Engineering
A reproducible path from data to a running service
- Versioned boundariesData assets, environments, components, models, and container tags are identifiable and reproducible.
- Cloud executionPreprocessing and model training run on created Azure ML resources rather than only on a local workstation.
- Deployment checksThe workflow verifies service health and a test prediction and reports pipeline status to Discord.
- ExplainabilityThe Streamlit dashboard uses Grad-CAM to show which image regions influenced a prediction.
- Operational manifestsKubernetes definitions include probes, resource limits, persistence, secrets, and multiple API replicas.
- Precise scopeAzure ML training and registration were executed in the cloud; Kubernetes was delivered as deployment manifests, not presented as a live production cluster.