Course Syllabus

  Module 1 (4 hours): Introduction to MLOps and Azure

• Description: Covers MLOps fundamentals and Azure ML basics. Learners explore DevOps principles applied to ML (source control, CI/CD, reproducible pipelines). The Azure ecosystem is introduced: Azure Machine Learning workspace, compute targets, and Azure DevOps/GitHub for code.
• Topics Covered: MLOps overview, DevOps principles for ML, Azure ML workspace setup, Azure subscription and resource organization, Git basics.
• Hands-on Lab: Set up an Azure ML workspace and a Git repository; connect Azure ML to Azure DevOps or GitHub.

  Module 2 (3 hours): Version Control and Collaboration

• Description: Focuses on source control and collaborative workflows. Students learn Git branching strategies (feature branches, pull requests) and automated testing linting for ML code. Versioning of experiments and datasets is introduced (using Azure ML Git integration or MLflow).
• Topics Covered: Git/GitHub/Azure Repos, branching and merging, code reviews, unit testing basics, tracking experiments in code.
• Hands-on Lab: Practice Git workflows: create branches, commit code changes for a model script, and merge to a main branch.

  Module 3 (4 hours): Azure Machine Learning Fundamentals

• Description: Introduces core Azure ML concepts. Learners use Azure ML Studio and SDK to create compute resources (clusters, instances) and register datasets. The concept of Azure ML Environments (reproducible software dependencies) is covered. Students run simple training experiments and view logs/metrics.
• Topics Covered: Azure ML workspace & compute, Datastores and Datasets for data versioning, Environments (Docker/conda), running experiments (training runs), logging metrics.
• Hands-on Lab: Train a sample model (e.g. Iris classifier) on Azure compute, log its metrics, and explore the run history.

  Module 4 (4 hours): Data Preparation and Versioning

• Description: Focuses on data management in Azure. Topics include ingesting and versioning data via Azure ML Dataset and DataStore, and basic ETL pipelines (e.g. with Azure Data Factory or Python). Learners also get an introduction to data drift – monitoring how production data may shift over time.
• Topics Covered: Azure Blob Storage/Data Lake, Azure ML Datasets/DataStores, feature engineering, data labeling, introduction to data profiling and drift.
• Hands-on Lab: Import a real dataset into Azure ML (e.g. CSV to DataStore), register it as a Dataset, and explore profiling tools.

  Module 5 (5 hours): Experimentation and ML Pipelines

• Description: Covers building reusable ML pipelines. Students learn to orchestrate training workflows by chaining steps (data prep, training, validation) in Azure ML Pipelines. They practice parameterizing pipelines for different data or model settings. This ensures experiments are reproducible and traceable.
• Topics Covered: Azure ML Pipeline creation (Python SDK or Designer), pipeline components (preprocessing, training, evaluation), parameterization, pipeline cloning and versioning.
• Hands-on Lab: Develop an Azure ML Pipeline for model training (including data preprocessing step), run it end-to-end, and examine outputs and metrics.

  Module 6 (4 hours): Model Registration and Packaging

• Description: Teaches how to register and package models. After training, models are stored in the Azure ML Model Registry. Learners practice registering multiple versions with metadata. The module also covers packaging a model with its environment and creating a Docker image. This image is stored in Azure Container Registry for deployment.
• Topics Covered: Model registration and versioning in Azure ML, model metadata, building inference environments (conda/pip), creating and pushing Docker images to ACR.
• Hands-on Lab: Register a trained model in Azure ML and examine its versioning. Package the model into a scoring image and push to Azure Container Registry.

  Module 7 (5 hours): Model Deployment

• Description: Explores deploying models as web services. Students deploy models to Azure Container Instances (ACI) for testing and to Azure Kubernetes Service (AKS) for scalable production. They write scoring (entry) scripts and configure deployment resources. Concepts of real-time vs batch endpoints are covered. Deployment also introduces continuous deployment – e.g. updating a running service with a new model version.
• Topics Covered: Deploy to ACI/AKS, create/manage endpoints (real-time, batch), traffic splitting (canary), environment inference configuration, using Application Insights for live metrics.
• Hands-on Lab: Deploy the registered model as an online endpoint on AKS. Test it by calling the endpoint and observe logs/metrics in Azure Monitor.

  Module 8 (3 hours): Monitoring, Drift Detection, and Retraining

• Description: Focuses on post-deployment monitoring. Learners configure Azure Monitor/Application Insights to track prediction performance and health. They set up data and model drift detection, and understand alerts on key ML events (e.g. continuous drift). The module covers how to use alerts to trigger retraining workflows (e.g. via Azure ML Pipelines and Azure Functions). Best practices for ML model reliability are emphasized.
• Topics Covered: Azure Monitor metrics/logs for endpoints, Application Insights integration, creating alerts on data drift or performance degradation, automated retraining triggers.
• Hands-on Lab: Enable data drift monitoring on the production endpoint. Simulate drift by feeding new data and configure an alert that triggers a retraining pipeline.

  Module 9 (4 hours): CI/CD and Automation with Azure DevOps

• Description: Teaches building end-to-end CI/CD pipelines using Azure Pipelines or GitHub Actions. Students automate workflows: on code or data changes, the pipeline retrains the model, tests it, and deploys if successful. Topics include YAML pipelines, build vs release pipelines, branching strategies, and integrating with Azure ML. This ensures continuous delivery of ML updates.
• Topics Covered: Azure Pipelines fundamentals (build and release), YAML pipeline for ML (training and deployment jobs), triggers (Git commit, schedule), artifact management, approval gates.
• Hands-on Lab: Create an Azure DevOps pipeline that runs on a Git commit: it will trigger an Azure ML pipeline to train and update the deployed model.

  Module 10 (4 hours): Capstone Project

• Description: A comprehensive project that ties all modules together. Teams implement an end-to-end MLOps workflow on a chosen real dataset (e.g. time-series forecasting or classification). This includes versioning code/data, training multiple model iterations, deploying models, and setting up monitoring/alerts. Students document their pipeline and present the solution, demonstrating the use of Azure MLOps services throughout.
• Topics Covered: End-to-end ML solution, integrating all learned components (source control, pipelines, deployments, monitoring). Emphasis on best practices learned in earlier modules.
• Hands-on Project: Deliver a fully automated ML deployment pipeline in Azure, complete with a retraining trigger on data drift.