Learning path
ML Engineering
You turn models into reliable products: features, training, serving and monitoring.
Why this path
Classic machine learning still runs most predictions in business: churn, fraud, demand, pricing. ML engineers make those models repeatable, served, monitored and retrained, which is also the foundation for LLM work.
How to use this page: 1. learn the key concepts, which are the same everywhere; 2. pick your platform; 3. work through Crawl, Walk and Run.
Read alongside: The AI Universe · Certification prep on this site.
Step 1 · the same everywhere
Key concepts
Learn these ideas first. They are the same on every cloud and with every tool; only the product names change. Each one is a short page: what it is, why it matters, an example and the common mistake.
Learning basics
Features and data
Production ML
Step 2
Where do you work?
Pick your platform and you get one page with that platform's products in every step.
Beginner · 8 weeks
Crawl: You can build one thing end to end, with help.
Train a model honestly, track it and serve it.
Key ideas at this level, the same on every platform:
Open this level on: AWS · Azure · Google Cloud · Databricks · Open source
Learn: the concept, and the product on each platform
| Learn | AWS | Azure | Google Cloud | Databricks | Open source |
|---|---|---|---|---|---|
| Python for datapandas or Polars, plotting, notebooks | SageMaker Studio | Azure ML studio | Vertex AI Workbench | Databricks notebooks | Jupyter |
| Core MLtrain, validate, test; leakage; a dumb baseline first | SageMaker training jobs | Azure ML jobs | Vertex AI training | Databricks ML runtime | scikit-learn, PyTorch |
| Track experimentsevery run, its data and its score | SageMaker MLflow | Azure ML (MLflow) | Vertex AI Experiments | MLflow | MLflow |
| Serve ita simple prediction endpoint | SageMaker endpoints | Azure ML online endpoints | Vertex AI endpoints | Model Serving | FastAPI or BentoML |
Build on GitHub
Churn model, end to end
A tabular model from raw data to a tracked experiment to a served endpoint, with an honest baseline first.
You are ready to move on when
- Your model beats a simple baseline, and you say by how much.
- Anyone can re-run your best experiment.
- Your endpoint answers in under a second.
Intermediate · 3 months
Walk: You can build it so it keeps working, without help.
Make it production-grade: features, registry, pipelines and safe releases.
Key ideas at this level, the same on every platform:
Open this level on: AWS · Azure · Google Cloud · Databricks · Open source
Learn: the concept, and the product on each platform
| Learn | AWS | Azure | Google Cloud | Databricks | Open source |
|---|---|---|---|---|---|
| Feature storethe same features for training and serving | SageMaker Feature Store | Azure ML feature store | Vertex AI Feature Store | Feature engineering in Unity Catalog | Feast |
| Model registryversions, approvals, rollback | SageMaker Model Registry | Azure ML registry | Vertex AI Model Registry | Models in Unity Catalog | MLflow registry |
| Pipelinesretrain and test on a schedule | SageMaker Pipelines MWAA (Airflow) / Step Functions | Azure ML pipelines Data Factory / Fabric pipelines | Vertex AI Pipelines Cloud Composer (Airflow) | Lakeflow Jobs + MLflow Lakeflow Jobs | Metaflow or Kubeflow Airflow or Dagster |
| Safe releasesshadow, canary, rollback | SageMaker endpoints GitHub Actions + CodeDeploy | Azure ML online endpoints GitHub Actions or Azure Pipelines | Vertex AI endpoints GitHub Actions + Cloud Build | Model Serving GitHub Actions + bundles | FastAPI or BentoML GitHub Actions |
Build on GitHub
Features and serving
Same features for training and live predictions, a model registry, and a canary release with rollback.
You are ready to move on when
- Training and serving features match, and a test proves it.
- You can roll back a model in one step.
- Retraining runs without you.
Advanced · 6 months +
Run: You can design it, prove it and lead others through it.
Run models as a platform: monitoring, drift, cost and many teams.
Key ideas at this level, the same on every platform:
Open this level on: AWS · Azure · Google Cloud · Databricks · Open source
Learn: the concept, and the product on each platform
| Learn | AWS | Azure | Google Cloud | Databricks | Open source |
|---|---|---|---|---|---|
| Monitoring and driftnotice quiet decay; alert on it | SageMaker Model Monitor | Azure ML model monitoring | Vertex AI Model Monitoring | Lakehouse Monitoring | Evidently |
| Retraining policyrelease only when it beats the current model | SageMaker Pipelines | Azure ML pipelines | Vertex AI Pipelines | Lakeflow Jobs + MLflow | Metaflow or Kubeflow |
| Platform for teamstemplates, shared features, cost per model | Terraform or AWS CDK Cost Explorer + CUR | Bicep or Terraform Cost Management | Terraform Billing export to BigQuery | Declarative Automation Bundles (Asset Bundles) Billing system tables | Docker Compose Query plans and run times |
| From ML to LLMwhere classic models still win, and how they combine | Amazon Bedrock Bedrock evaluations | Azure OpenAI in Microsoft Foundry Foundry evaluations | Gemini on Vertex AI Vertex AI Gen AI evaluation | Foundation Model APIs MLflow evaluation | Ollama, or a free hosted open model Eval Harness Framework or promptfoo |
Build on GitHub
Drift and retrain loop
Watch a live model, detect drift, retrain on a schedule, and only release when it beats the current one.
Robotics data benchmark
A benchmark for the data pipelines that feed robot learning: formats, quality and cost.
You are ready to move on when
- Drift raises an alert before a user notices.
- Each model's monthly cost is known.
- A new team can ship a model with your template.