MLA-C02 study plan: pass in 6 weeks

Six weeks at 8–10 hours a week, assuming you meet AWS’s expectation of roughly a year with SageMaker AI and Bedrock. If you do not, see the prerequisites — this is not an exam you can read your way through.

You need an AWS account you can build in. Every domain here is about doing something, and the questions are written by people who assume you have.

Weeks 1 and 2: data preparation for ML and AI (28%)

The largest domain, and the one engineers coming from a modelling background underrate.

Week 1 — ingestion and transformation:

• Ingesting data from the common AWS sources, and the storage choices that follow.
• Transformation and cleaning at scale; common data formats and when each is appropriate.
• Feature engineering concepts and how features are stored and reused.

Week 2 — validation and readiness:

• Data validation: completeness, schema conformity, detecting corrupt or anomalous data.
• Splitting data for training, validation and test, and why leakage matters.
• Labelling workflows.
• Preparing data specifically for foundation model work, including content for retrieval rather than training.

Week 3: model and foundation model development (24%)

• Selecting a modelling approach for a stated problem.
• Training, and hyperparameter tuning — what the common parameters do and how tuning is run at scale.
• Analysing model performance: the right metric for the problem, and why accuracy alone misleads on imbalanced data.
• Model versioning and tracking experiments.
• Working with foundation models: when to prompt, when to retrieve, when to fine-tune.

Week 4: deployment and orchestration (24%)

• Choosing deployment infrastructure and endpoint types for a latency and cost requirement.
• Provisioning compute and configuring auto scaling.
• Real-time versus batch versus asynchronous inference, and which a scenario calls for.
• CI/CD pipelines for ML, and infrastructure as code.
• Agentic workflows — new in C02, and explicitly in AWS’s task list.

Week 5: operating, monitoring and securing (24%)

• Monitoring models, data and infrastructure; detecting drift and degradation.
• Observability for agentic workflows, also called out explicitly by AWS.
• Cost optimisation — a recurring theme, since AWS cares that you pick the cheapest option that meets the requirement.
• Logging and troubleshooting ML systems.
• Security: IAM and least privilege, encryption, data protection, and compliance features.

Week 6: practice and repair

• Full practice exam under real conditions. If you are sitting the beta form, that is 170 minutes for 85 questions.
• Sort mistakes by domain. With four domains at 24–28%, a single weak area is expensive but survivable under compensatory scoring.
• Rebuild the weakest, then take a second practice exam.

Where the hours go

The instinct AWS rewards

Across every domain, the expected answer is usually the cheapest managed option that meets the stated requirement.

If a scenario mentions cost, latency, or operational overhead, that clause is doing the work. A technically excellent answer that over-provisions is wrong. Practise reading for the constraint before reading the options — it eliminates two of four answers more often than any other technique.

Build one thing end to end

If you do nothing else, do this: take one dataset, prepare it, train a model, deploy it to an endpoint, put a CI/CD pipeline around it, and set up monitoring. One small project touching all four domains teaches more than any amount of reading, and it maps directly onto how the questions are framed.

Because scoring is compensatory, coverage matters more than mastery. Getting adequately through all four domains passes; excelling at two does not.