MLA-C02 study resources

MLA-C02 is three-quarters an operations exam. Reading covers the vocabulary; only production experience covers the judgement. If you have deployed a model, watched it drift and debugged why serving predictions differed from validation, you are most of the way there. If you have not, budget for building rather than reading.

Start with the official material

The exam guide. Four domains at near-equal weight: data preparation (28%), model and foundation model development (24%), deployment and orchestration (24%), and operating, monitoring and securing (24%). Only a quarter is model building. Let that shape your study.

AWS Skill Builder, specifically the machine learning engineer content. Free tier material is substantial here.

The SageMaker and MLOps documentation. This exam asks about behaviour and trade-offs that summary courses flatten — batch transform versus real-time endpoints, execution role scoping, autoscaling bounds.

Build the things the exam asks about

Each of these maps to recurring exam themes:

• A training job with a scoped execution role that can read one prefix and nothing else. You will remember least privilege after fighting it once.
• A batch inference job over a large dataset, then a real-time endpoint for the same model. The cost and latency difference becomes obvious rather than theoretical.
• A drift check: compute input distribution statistics at training, then compare against production inputs. This is the monitoring question the exam keeps asking.
• A pipeline that retrains on new data with the model registered and versioned. Reproducibility stops being abstract.
• A canary deployment with weighted traffic, then roll it back.

Worth paying for

Practice tests with explanations, for diagnosis across four evenly weighted domains. Start with the free 20-question test here.

Compute credits for the labs above, which is the genuinely valuable purchase for this exam. A modest amount goes a long way, and it buys the experience the questions test.

What to skip

Braindumps. Terms violation, revocation risk, and particularly useless here — the questions describe production situations rather than reciting facts.

Deep algorithm theory. You will not derive backpropagation. You will be asked whether recall or precision matters when missing a positive case is costly.

Material for the retired ML Specialty exam. Different exam, different emphasis, and considerably more theory-heavy than MLA-C02.

Courses that spend most of their time on model training. That is 24% of the exam. If a course spends 70% of its runtime there, it is mis-weighted.

A sensible order

1. Read the exam guide. Note that three of four domains are operations, not modelling.
2. Work through the Skill Builder engineer content.
3. Build the five things above. Break each deliberately.
4. Take the practice test.
5. Return to documentation for the weakest domain.

Is this the right exam?

If you build and operate models, yes. If you work near ML without doing it, AIF-C01 is the right level — see the AWS AI certification path. For the cross-vendor question, MLA-C02 vs AI-103 explains why these two associate exams point in different directions.