Course Syllabus

Module 1: Introduction to MLOps & GCP (2h)

• Topics: MLOps lifecycle and best practices; overview of Vertex AI, Kubeflow, BigQuery ML, and Cloud ML services.
• Benefits: Understand how MLOps shortens development cycles and improves reliability.
• Lab/Exercise: Setup GCP project; create a Vertex AI Workbench notebook; run a simple sample training and deployment.
• Key takeaway: Grasp MLOps concepts and the role of Google Cloud’s ML platform in automating the model development and deployment process.

Module 2: Data Infrastructure for ML (4h)

• Topics: Google Cloud Storage for data/model artifact storage; BigQuery for large-scale data analysis and SQL-based ML (BigQuery ML).
• Tools: Loading and querying data in BigQuery; using the BigQuery ML CREATE MODEL syntax for a quick regression/classification model.
• Lab: Ingest a dataset into Cloud Storage and BigQuery; train a simple BigQuery ML model (e.g. linear regression) and evaluate results.
• Key takeaway: Learn how to build repeatable data pipelines and leverage BigQuery ML as part of an MLOps workflow . (BigQuery ML models can be a staging point in Vertex pipelines.)

Module 3: Vertex AI Workbench & Tools (4h)

• Topics: Vertex AI Workbench (JupyterLab) environments; Vertex AI Python SDK; integration with other services.
• Tools: Create and use a managed workbench instance; access Cloud Storage and BigQuery from notebooks; use Vertex AI SDK to launch jobs.
• Lab: Develop a small model training script in Vertex Workbench, save data to Cloud Storage, and submit a Vertex AI custom training job.
• Key takeaway: Gain hands-on familiarity with Vertex AI development environment and how code in notebooks maps to cloud execution.

Module 4: Model Training on Vertex AI (5h)

• Topics: Custom model training vs AutoML; setting up TensorFlow/PyTorch jobs on Vertex AI; hyperparameter tuning.
• Tools: Vertex AI Custom Training, built-in algorithms, AutoML Vision/Tabular/NLP, and hyperparameter tuning jobs.
• Lab: Train a custom image or tabular model on Vertex AI; run a hyperparameter tuning job to optimize performance.
• Key takeaway: Learn to train scalable models on managed infrastructure; reduce training time with built-in hyperparameter tuning features.

Module 5: Building ML Pipelines with Kubeflow (6h)

• Topics: Vertex AI Pipelines (Kubeflow Pipelines) architecture ; defining pipeline components (prebuilt vs custom).
• Tools: Kubeflow Pipelines SDK (v2) with Google Cloud Pipeline Components; use Vertex TensorBoard and Vertex Experiments.
• Lab: Create an end-to-end Kubeflow pipeline: data preprocessing, model training, evaluation, and deployment steps. Use Google’s pipeline components for common tasks.
• Key takeaway: Orchestrate reproducible ML workflows as DAGs of containerized tasks, enabling automation and tracking of the entire model lifecycle  .

Module 6: CI/CD for ML with Cloud Build (5h)

• Topics: Continuous integration/continuous delivery (CI/CD) for ML; version control with Cloud Source Repositories; containerizing models.
• Tools: Cloud Build pipelines, Artifact Registry for Docker images, Cloud Container Registry (optional), Git triggers.
• Lab: Set up a CI pipeline: on code commit, Cloud Build packages a model training app into a container, pushes it to Artifact Registry, and optionally triggers a Vertex AI deployment.
• Key takeaway: Automate testing and deployment of ML code. Example: use Cloud Build + Artifact Registry to build and push a Docker image of an ML service .

Module 7: Event-Driven Automation (3h)

• Topics: Triggering pipelines on events (data arrival or code push) using Cloud Functions and Cloud Scheduler.
• Tools: Cloud Functions (Python/Node.js) to invoke Vertex AI API; Cloud Pub/Sub or Eventarc for messaging; Cloud Scheduler for cron-like triggers.
• Lab: Implement a Cloud Function that launches a Vertex AI Pipeline when new data appears in a Cloud Storage bucket.
• Key takeaway: Enable automation by responding to events. For instance, a Cloud Function can automatically start a training pipeline on data update .

Module 8: Feature Store & Model Registry (2h)

• Topics: Vertex AI Feature Store for centralized feature management; Vertex AI Model Registry for tracking model versions.
• Tools: Create online/offline feature stores, ingest features; register trained models to Vertex Model Registry.
• Lab: Store ML features in Vertex Feature Store; serve them in batch or real-time. Register a model in Vertex and view metadata.
• Key takeaway: Learn to manage features at scale and organize models. (Wayfair example: using Vertex Feature Store to serve features in real time without manual infrastructure management .)

Module 9: Monitoring & Logging (2h)

• Topics: Vertex AI Model Monitoring (data drift and skew detection) ; Cloud Monitoring and Logging for pipelines.
• Tools: Configure Model Monitoring alerts on deployed endpoints; use Cloud Logging to view Vertex Pipeline logs; dashboards in Cloud Monitoring.
• Lab: Deploy a model to Vertex AI Endpoint and enable model monitoring; simulate data skew and observe alerts.
• Key takeaway: Ensure model quality in production. Vertex AI will alert if inference data drifts from training distribution . System-level metrics (CPU, errors) are tracked via Cloud Monitoring.

Module 10: Real-World Case Studies (3h)

• Topics: Analyze industry MLOps implementations.
• Case Study – Wayfair (Retail): Built a cloud-native MLOps platform on Vertex AI. Shifted from legacy Airflow to Vertex Pipelines + Feature Store; standardized CI/CD and reduced hyperparameter tuning from two weeks to one day  .
• Case Study – Vodafone (Telecom): Developed the “AI Booster” on Google Cloud with Vertex AI, Cloud Build, and Cloud Functions. Automated thousands of model pipelines; cut PoC-to-production time by ~80% (from months to ~4 weeks)  .
• Lab/Discussion: Walk through a simplified version of one of these pipelines; identify how services (Vertex AI Pipelines, Feature Store, Cloud Build, Functions) integrate.
• Key takeaway: Learn proven design patterns. For example, Vodafone’s platform uses Cloud Build and Functions to automate deployment , achieving faster, more reliable ML releases .

Module 11: Capstone Project – End-to-End MLOps Pipeline (3h)

• Project: Work in small teams on an industry-agnostic dataset (e.g., customer churn or predictive maintenance).
• Tasks: Build the full MLOps workflow: data ingestion (Cloud Storage/BigQuery), model training (Vertex AI), pipeline orchestration (Vertex AI Pipelines), CI/CD (Cloud Build), deployment (Vertex Endpoints), and monitoring.
• Support: Leverage all tools learned; instructors guide integration.
• Key takeaway: Apply concepts hands-on to cement learning. Deliver a working end-to-end pipeline that can be demonstrated.

Module 12: Certification Prep & Review (1h)

• Topics: Recap key services and workflows relevant to the Professional ML Engineer exam: architecture, data management, modeling, optimization, and ML Ops.
• Activities: Quiz-style review questions covering module content (pipelines, Feature Store, Cloud Build, etc.). Discuss exam tips.
• Key takeaway: Solidify understanding of GCP MLOps best practices and ensure readiness for certification objectives.