MLA-C02 study scenarios and example profiles

How to study for MLA-C02 depends on which half of “ML engineer” you already do. The exam is flat across four domains: Data Preparation for ML and AI at 28%, and model development, deployment and orchestration, and operating, monitoring and securing at 24% each. The passing score is 720 on a 100 to 1,000 scale, scoring is compensatory, and AWS expects about a year of SageMaker AI and Bedrock experience.

Example profiles, not real people. The profiles below are illustrative planning examples built from the official exam outline. They are not reviews or reports of anyone’s result.

The site’s six-week study plan assumes 8 to 10 hours a week, roughly 48 to 60 hours, with about 16 to 20 hours on data preparation and 8 to 10 on each other domain. The budgets here are planning assumptions that move hours between those blocks. They do not guarantee a result.

Profile one: the data scientist trained on notebooks

Starting point. This candidate picks algorithms, tunes hyperparameters and evaluates models with the right metric. Model development (24%) is comfortable. Pipelines, endpoints and infrastructure as code are somebody else’s job.

Where the time goes. Deployment and orchestration (24%): endpoint types, auto scaling, real-time against batch and asynchronous inference, CI/CD for ML and agentic workflows. Then operating, monitoring and securing (24%): drift detection, observability, IAM least privilege and encryption.

The trap. Spending weeks on model theory. AWS lists deep work in two or more ML domains as out of scope, so extra modelling depth earns little, while a deployment gap costs marks in two domains. The deployment guide is the place to start.

Suggested order. Weeks four and five of the plan first, then data preparation, with model development as a light refresh.

Time budget. A reasonable planning assumption is 14 hours on deployment, 12 on operating and securing, 14 on data preparation, 4 on model development, and a full practice week.

Profile two: the DevOps or backend engineer new to ML

Starting point. This candidate already writes CI/CD pipelines, provisions infrastructure as code, reads CloudWatch and understands IAM. Machine learning concepts and generative AI vocabulary are new.

Where the time goes. Data preparation (28%) and model development (24%) together hold more than half the marks and contain most unfamiliar terms: feature engineering, class imbalance, data leakage, hyperparameter tuning, retrieval, chunking and fine-tuning strategies. The data preparation guide and model development guide cover them.

The trap. Trusting platform skill to carry the exam. Questions ask which metric fits an imbalanced dataset or when to prompt, retrieve or fine-tune, which no pipeline experience answers. The prerequisites page also puts missing hands-on SageMaker AI or Bedrock use first among gaps.

Suggested order. Build one small training and deployment project first, then weeks one to three of the plan, then monitoring.

Time budget. A reasonable planning assumption is 20 hours on data preparation, 14 on model development, 6 on deployment as a quick check, and the rest on monitoring and practice. If under a year of experience, lengthen the timeline rather than compress it.

Profile three: the data engineer who feeds the models

Starting point. This candidate builds ingestion, Glue jobs, streaming with Kinesis and storage formats, so the 28% domain is largely familiar. SageMaker Feature Store and vector databases may be new.

Where the time goes. Model development and evaluation: choosing between Bedrock models, tuning and early stopping, NLP metrics, MLflow, LLM-as-a-judge and RAG retrieval accuracy. In operating and securing, model-specific monitoring such as drift.

The trap. Skipping the largest domain because it feels known. It still includes embeddings, RAG document preparation, bias handling and data validation specific to training data, which ordinary pipelines rarely cover.

Suggested order. A short data preparation audit against the outline, then model development, deployment and monitoring.

Time budget. A reasonable planning assumption is 8 hours on the data domain audit, 14 on model development, 10 on deployment, 10 on operating and securing, and a full practice week.

MLA-C02 profiles side by side

How to check which MLA-C02 profile fits before booking

Run the five hands-on checks on the prerequisites page: deploy to an endpoint, run a training job, wrap a pipeline, set up monitoring and try the free sample questions. Sort the misses by domain, then take a timed practice test and review the format and scoring in the exam format page. If your work is using AI on AWS rather than building it, AIF-C01 is the exam that fits.