How to pass MLA-C02: a 6-week AWS ML Engineer study plan

AWS recommends at least one year with SageMaker AI, Bedrock and other ML services, and experience with both traditional ML and generative AI. It also expects knowledge of common algorithms, data formats and ingestion, querying, software engineering practices, CI/CD and infrastructure as code.

If part of that foundation is missing, stretch the plan to eight weeks or start with the digital courses in the Exam Prep Plan. If you have never worked with AI, AIF-C01 gives you the concepts first.

Week 1 (domain 1, 28%): ingestion from S3, RDS, DynamoDB and streaming with Kinesis; Parquet, ORC, JSON and CSV formats; transformations with AWS Glue, DataBrew and SageMaker Data Wrangler; Feature Store, data quality, bias, class imbalance and document preparation for RAG. Week 2 (domain 2, classic ML): SageMaker AI built-in algorithms, script mode, automatic model tuning, early stopping, distributed training, overfitting and explainability.

Week 3 (domain 2, generative AI): choosing foundation models in Bedrock, fine-tuning versus prompt engineering, RAG patterns, evaluation with metrics such as BLEU, ROUGE and BERTScore, LLM-as-a-judge and Bedrock evaluations. Week 4 (domain 3): real-time and batch inference, endpoints in a VPC, auto scaling, containers, knowledge bases, agents, Model Registry, Prompt Management and pipelines with CodePipeline.

Week 5 (domain 4): monitoring with CloudWatch, data drift, A/B testing, AgentCore Observability, inference and token costs, purchasing options, least-privilege IAM, CloudTrail, AWS Config and Amazon Bedrock Guardrails. Week 6: full mock exams, review of mistakes by domain and the list of short service names AWS provides through the Help button in the exam.

Your base is the MLA-C02 exam guide on docs.aws.amazon.com: it lists every task and skill, plus in-scope and out-of-scope services. Use it as a checklist and tick off what you already know.

AWS offers a four-step Exam Prep Plan on Skill Builder: get to know the format with the Official Practice Question Set and the Official Pretest, fill gaps with digital courses and AWS Builder Labs, AWS Cloud Quest and AWS Jam, review with exam-style questions, flashcards and AWS SimuLearn, and check readiness with the Official Practice Exam. Check on Skill Builder which parts need a subscription.

Build a small project end to end: prepare data with Glue or Data Wrangler, train and tune a model in SageMaker AI, register it in Model Registry and deploy it to an endpoint inside a VPC with auto scaling. Add a CloudWatch alarm and a drift check.

In parallel, create a Bedrock knowledge base over your own documents, try two chunking strategies, apply Guardrails and measure answer quality with a Bedrock evaluation. Those two exercises touch skills in all four domains and show you the real cost of each choice.

Start with the MLA-C02 sample questions to see the style, and log your mistakes by domain, not just your score. Then practice by topic in the MLA-C02 question bank (start free) and go back to the explained outline whenever a topic gives you trouble.

From week 5, take timed full mock exams. In the standard version the passing score is 720 out of 1,000 and you do not need to pass each section separately, but read every scenario carefully: cost, latency or security requirements usually decide the correct answer.

Studying only generative AI because it is new, when data preparation weighs 28% and classic ML is still in the outline. Another common gap is domain 4: costs, quotas, credentials for model access and IAM show up across many skills.

On exam day, the classic mistake is leaving questions blank: they count as wrong and there is no penalty for answering. And in the beta, sitting the exam before you are ready is costly, because you can take it only once and must wait for the standard version to retake.