
AWS Certified Machine Learning Engineer – Associate checks that you can prepare data, train and evaluate models, deploy them, and run them safely on AWS. The exam has moved to a new version, MLA-C02, which keeps the same four domains but adds a lot of generative AI: vector databases, embeddings, RAG, Amazon Bedrock knowledge bases, agents, prompt management, LLM evaluation and Bedrock Guardrails. MLA-C02 has been in beta since 29 September 2026; the English MLA-C01 exam ended on 28 September 2026.
This guide follows the official AWS exam page and the MLA-C02 exam guide: exam facts, the four domains with weights, what changed from C01, the AWS services to know, a 6-week plan and 15 original practice questions.
Quick facts (AWS, November 2026)
| Item | Detail |
|---|---|
| Exam | AWS Certified Machine Learning Engineer – Associate (MLA-C02), English beta |
| Beta exam | 85 questions, 170 minutes, US$75 (beta pricing); beta results arrive after the beta closes |
| Standard exam (per exam guide) | 50 scored and 15 unscored questions; multiple choice and multiple response |
| Passing score | 720 on a 100–1,000 scale |
| Validity | 3 years |
| Recommended experience | At least 1 year with SageMaker AI, Amazon Bedrock and related AWS services, plus 1 year in a role such as backend developer, DevOps, data engineer or data scientist |
| Other languages | Korean, Japanese and Simplified Chinese MLA-C01 stay available until MLA-C02 reaches general availability |
Check the official exam page and the MLA-C02 exam guide for price and availability when you book.
Domains and weights
| Domain | MLA-C01 | MLA-C02 | Tasks |
|---|---|---|---|
| 1. Data Preparation for ML and AI | 28% | 28% | Collect and store data; transform data, engineer features and pre-process; validate data quality and manage bias |
| 2. ML Model and Foundation Model (FM) Development | 26% | 24% | Choose modelling approaches; train, fine-tune and customise models; analyse and evaluate ML and AI systems |
| 3. Deployment and Orchestration of ML and AI Workflows | 22% | 24% | Manage deployment infrastructure; provision resources; automate orchestration and CI/CD for MLOps and AI |
| 4. Operating, Monitoring, and Securing ML and AI Solutions | 24% | 24% | Monitor inference and performance; optimise cost and performance; secure workloads and endpoints |
What is new in MLA-C02
- Data for GenAI: vector stores (OpenSearch Service, RDS with pgvector, Amazon S3), embedding models, tokenisation, chunking documents for RAG, masking and redacting data, and preparing data for fine-tuning, continued pre-training and distillation.
- Model choice: picking a foundation model in Amazon Bedrock, fine-tuning strategies, RAG patterns, and the trade-off between quality, latency and cost.
- Evaluation: human-in-the-loop review, BLEU, ROUGE, BERTScore, semantic similarity, LLM-as-a-judge, and retrieval accuracy for RAG.
- Deployment: FM hosting options, importing outside models (Bedrock Custom Model Import), Bedrock knowledge bases, agents and agent state, GPU scaling, prompt management and prompt testing.
- Operations and security: monitoring agents and tool failures, token and embedding costs, Bedrock API keys vs IAM credentials, Bedrock Guardrails, and scanning pipelines with Amazon Inspector and CodeGuru.
- Removed: SageMaker Neo for edge, bring-your-own-container on SageMaker, model-size reduction, and loading training data from EFS or FSx.
Services to know
| Area | Services |
|---|---|
| Data | Amazon S3, AWS Glue, Glue DataBrew, Glue Data Quality, Amazon EMR (Spark), Kinesis, Amazon MSK, SageMaker Data Wrangler, SageMaker Feature Store, SageMaker Ground Truth |
| Models | SageMaker AI training, built-in algorithms, automatic model tuning, JumpStart, Clarify (bias and explainability), Model Registry; Amazon Bedrock models, fine-tuning and evaluations |
| Deployment | SageMaker endpoints (real-time, serverless, asynchronous, batch transform), auto scaling, SageMaker Pipelines, Bedrock knowledge bases and agents, CodePipeline, CloudFormation and CDK |
| Operations | SageMaker Model Monitor, CloudWatch, CloudTrail, Cost Explorer, IAM, KMS, VPC endpoints, Amazon Inspector, Bedrock Guardrails |
6-week study plan
| Week | Focus | Lab |
|---|---|---|
| 1 | Domain 1: storage choices, formats (Parquet, CSV, JSON), Glue and DataBrew, streaming, Feature Store | Clean a CSV with DataBrew, write Parquet to S3 and register features |
| 2 | Domain 1 for GenAI: embeddings, chunking, vector stores, bias with Clarify, class imbalance, PII masking | Chunk 20 PDFs, create embeddings and load them into OpenSearch or pgvector |
| 3 | Domain 2: algorithm choice, training, tuning, metrics (precision, recall, F1, RMSE, AUC), Model Registry | Train XGBoost in SageMaker and run a hyperparameter tuning job |
| 4 | Domain 2 for GenAI: Bedrock model choice, prompt engineering, fine-tuning, RAG design, LLM evaluation | Build a Bedrock knowledge base and compare two models with Bedrock evaluations |
| 5 | Domain 3: endpoint types, auto scaling, SageMaker Pipelines, CI/CD, agents, prompt management | Deploy a model to a serverless endpoint; automate retraining with Pipelines |
| 6 | Domain 4: Model Monitor, CloudWatch, cost control, IAM, KMS, VPC, Guardrails; review | Add a Guardrail and data-capture monitoring; take the official practice question set; book the exam |
15 practice questions with answers
These are original questions written for this guide, not exam questions.
1. Which file format is best for large analytical training datasets read column by column?
Apache Parquet.
2. Which service stores features for reuse in both training and online inference?
SageMaker Feature Store.
3. Which service detects bias in training data and explains predictions?
SageMaker Clarify.
4. A dataset has 2% positive cases. Name two ways to handle the imbalance.
Oversample the minority class (for example SMOTE) or undersample the majority, or use class weights.
5. Fraud detection: which metric matters more if missed fraud is costly?
Recall.
6. Why split documents into chunks before embedding them for RAG?
Smaller chunks give more precise retrieval and fit the model's context window.
7. Name two AWS options for storing vectors.
Amazon OpenSearch Service and Amazon RDS for PostgreSQL with pgvector (Amazon S3 vectors is another).
8. Which Bedrock feature connects a model to your documents for RAG without building the pipeline yourself?
Amazon Bedrock Knowledge Bases.
9. Which metrics compare generated text with a reference summary?
ROUGE (also BLEU and BERTScore).
10. Traffic is rare and unpredictable, and a few seconds of cold start is fine. Which SageMaker inference option?
Serverless inference.
11. Predictions are needed for a 50 GB file once a night. Which option?
Batch transform.
12. Which service orchestrates data processing, training, evaluation and registration as one workflow?
SageMaker Pipelines.
13. Which tool detects data drift on a live endpoint?
SageMaker Model Monitor.
14. How do you block harmful content and PII in a Bedrock application?
Amazon Bedrock Guardrails.
15. How do you keep training traffic off the public internet?
Run jobs in a VPC with VPC endpoints for S3 and SageMaker, and encrypt data with KMS.
Before and after MLA
- New to AWS AI? Start with Cloud Practitioner or AI Practitioner.
- Go deeper in GenAI with the AWS Generative AI Developer Professional, or data with Data Engineer Associate.
- Learn it hands-on in our AWS Bedrock GenAI course and AI, GenAI and agentic AI course; see live ML and AI jobs.
Cloud Soft Solutions in Ameerpet, Hyderabad teaches AWS, machine learning, generative AI and MLOps with real-time projects. See the APEX program and all courses.
📞 Call +91 96660 19191 · WhatsApp · Flat No. 513, 5th Floor, Nilagiri Block, Aditya Enclave, Ameerpet, Hyderabad
Frequently Asked Questions
What is MLA-C02?
The new version of the AWS Certified Machine Learning Engineer – Associate exam, in English beta since 29 September 2026.
Can I still take MLA-C01?
Not in English; English MLA-C01 ended on 28 September 2026. Korean, Japanese and Simplified Chinese MLA-C01 remain until MLA-C02 reaches general availability.
How long is the MLA-C02 beta exam?
170 minutes for 85 questions, at beta pricing of US$75.
What is the MLA passing score?
720 on a scale of 100–1,000.
What is new in MLA-C02?
Generative AI skills: vector stores, embeddings, RAG, Bedrock knowledge bases and agents, LLM evaluation, prompt management and Bedrock Guardrails.
How long is the AWS ML Engineer Associate valid?
3 years.



