Full exam simulation
65 questions · 130 minutes, 15 unscored and not identified. One question per screen, flag and strike-out, a review screen and a score report per domain — no answers until you finish.
Amazon Web ServicesMLA-C02AssociateMachine LearningAvailable
AWS Certified Machine Learning Engineer – Associate
Build, deploy, operate and secure ML and generative AI solutions on AWS with SageMaker AI and Amazon Bedrock.
Independent practice — not affiliated with or endorsed by Amazon Web Services. MLA-C02 is used only to identify the exam. Trademark notices
MLA-C01 in English ended September 28, 2026; MLA-C02 is in beta (English only) and MLA-C01 stays available in Japanese, Korean and Simplified Chinese until MLA-C02 general availability.
Domain weights
Share of scored content per domain. Every sitting draws questions by these weights.
- D1Data Preparation for ML and AI28 %18 concepts
- D2ML Model and Foundation Model (FM) Development24 %16 concepts
- D3Deployment and Orchestration of ML and AI Workflows24 %16 concepts
- D4Operating, Monitoring, and Securing ML and AI Solutions24 %16 concepts
Your progress
No attempts yet. Take a full sitting to get a readiness rating; results are kept in this browser. Join your class to keep them on every device.
Choose a mode
ESL +30
The full sitting with 30 extra minutes (160 min) — the accommodation non-native English speakers can request for exams in English.
Quick 20
20 questions · 40 min, same weights, all scored. A daily check.
Weak spots
20 questions aimed at your weakest domains, missed concepts and due reviews.
Practice builder
Pick domains, topics, how many questions and which pool (new, wrong, due, low confidence). Spaced repetition with confidence ratings, flashcards and a mistake book.
Open practiceDomain drill
One domain, up to 20 questions.
What's on the exam
Domain 1: Data Preparation for ML and AI
28 %Collecting and storing data (formats, streaming, vector stores, Feature Store), transformation and feature engineering (Glue, Data Wrangler, embeddings, RAG chunking, masking) and data quality and bias.
Practice domain 1 →Domain 2: ML Model and Foundation Model (FM) Development
24 %Choosing modeling approaches (Bedrock FMs, built-in algorithms, AI services, RAG patterns), training, tuning and fine-tuning, and evaluating ML and GenAI systems (metrics, drift baselines, shadow tests, human review).
Practice domain 2 →Domain 3: Deployment and Orchestration of ML and AI Workflows
24 %Deployment infrastructure and inference options, provisioning endpoints, containers, knowledge bases and agents, and CI/CD and orchestration for MLOps (Pipelines, Model Registry, prompt and agent versioning).
Practice domain 3 →Domain 4: Operating, Monitoring, and Securing ML and AI Solutions
24 %Monitoring models and agents (Model Monitor, drift, A/B tests, observability), cost and performance optimization (instances, purchasing options, token costs) and security (IAM, VPC isolation, logging, Guardrails).
Practice domain 4 →
Domain names and weights come from the official MLA-C02 exam guide (checked 2026-10-26). Always check it for the authoritative scope.
Sources · checked 2026-10-26
Sample questions
Two of this exam's original questions, with the explanation you get after every answer.
Sample 1Select one
A training job reads a 2 TB dataset from S3 once per epoch, and start-up time spent downloading data is too long. Which input mode reduces start-up time without changing storage?
- ACopying the data to the notebook instance
- BStoring the data in DynamoDB
- CFastFile mode, which streams S3 objects on demand through a file system interface
- DFile mode with a larger EBS volume and a higher provisioned IOPS setting for the download
Show answer & explanation
Correct: C. FastFile mode exposes S3 objects as files and streams them as they are read, so training starts without downloading the full dataset.
Sample 2Select 2
Which transformations change numeric feature scale? (Select TWO.)
- AMin-max normalization to a 0-1 range
- BDeduplication
- COne-hot encoding
- DStandardization to zero mean and unit variance
- ETokenization
Show answer & explanation
Correct: A, D. Standardization and min-max scaling rescale numbers; one-hot encoding handles categories.