AWS · MLA-C02
AWS Certified Machine Learning Engineer - Associate (MLA-C02) Practice Tests
Validates expertise in designing, deploying, and operating machine learning solutions on AWS, including data preparation, model development, deployment/orchestration, and monitoring of ML workloads. Covers generative AI, Amazon Bedrock, and agentic AI topics.
Exam Details
Practice Questions
321
≈ 4 practice exams
Duration
170 minutes
Passing Score
720/1000
Difficulty
AssociateLast Updated
Sep 2026
Topics Covered
Data Preparation for ML and AIML Model and Foundation Model (FM) DevelopmentDeployment and Orchestration of ML and AI WorkflowsOperating, Monitoring, and Securing ML and AI Solutions
AWS Certified Machine Learning Engineer - Associate (MLA-C02) Practice Exam Preparation
Use this MLA-C02 practice exam to prepare for AWS Certified Machine Learning Engineer - Associate (MLA-C02) with realistic questions, detailed explanations, and focused study modes. The practice bank includes 321 questions for AWS MLA-C02, so you can review the exam steadily instead of relying on one long cram session.
As you practice, pay extra attention to recurring topics such as Data Preparation for ML and AI, ML Model and Foundation Model (FM) Development, Deployment and Orchestration of ML and AI Workflows, and Operating, Monitoring, and Securing ML and AI Solutions. Start with short sessions to identify weak areas, then move into timed quizzes once your accuracy is consistent.
The explanations are especially useful when you want to connect exam wording to the responsibilities and scenarios described in the official certification guidance. Use the free preview first, then unlock the full question bank when you are ready to build a complete study routine.
Exam Domain Breakdown
Content Domain 1: Data Preparation for ML and AI28%
Content Domain 2: ML Model and Foundation Model (FM) Development24%
Content Domain 3: Deployment and Orchestration of ML and AI Workflows24%
Content Domain 4: Operating, Monitoring, and Securing ML and AI Solutions24%
Exam Overview
The AWS Certified Machine Learning Engineer - Associate (MLA-C02) validates a candidate's ability to build, operationalize, deploy, and maintain AI and ML solutions and pipelines using the AWS Cloud. The exam covers both traditional ML workflows and modern generative AI (GenAI) workloads, including foundation models (FMs), Retrieval Augmented Generation (RAG) architectures, and agentic AI systems. Core AWS services tested include Amazon SageMaker AI, Amazon Bedrock, AWS Glue, Amazon Kinesis, and a range of storage and orchestration services.
MLA-C02 is an updated version of the MLA-C01 exam, expanding significantly into generative AI territory. New content areas include configuring vector databases for AI applications, preparing data for FM fine-tuning and continuous pre-training, deploying and managing Amazon Bedrock agents, implementing RAG pipelines, monitoring agentic workflows, and applying responsible AI safeguards using Amazon Bedrock Guardrails. This makes the certification distinctly relevant for engineers working at the intersection of MLOps and modern GenAI systems.
Who Should Take This Exam
This certification is designed for ML engineers, MLOps engineers, LLMOps engineers, data engineers, backend software developers, and data scientists who work with AWS-based ML and AI pipelines. The target candidate should have at least one year of hands-on experience with Amazon SageMaker AI, Amazon Bedrock, and related AWS ML services, as well as at least one year of experience in a related technical role such as backend development, DevOps, or data science.
Candidates should be comfortable with both traditional ML workflows (model training, hyperparameter tuning, deployment pipelines) and generative AI concepts (foundation models, prompt engineering, RAG, agentic workflows). This exam is not intended for individuals focused purely on ML research, data analysis, or full end-to-end solution architecture—those tasks are explicitly out of scope.
Prerequisites
AWS does not impose formal prerequisites for the MLA-C02 exam, but the recommended experience profile is substantial. Candidates should have at least one year of practical experience using SageMaker AI and Amazon Bedrock for ML engineering tasks. General IT knowledge requirements include familiarity with common ML algorithms and their use cases, understanding of FM capabilities and limitations, data engineering fundamentals (ingestion, transformation, common formats like Parquet, JSON, ORC), software engineering best practices, CI/CD pipelines, and infrastructure as code (IaC).
On the AWS side, candidates should be proficient with SageMaker AI algorithms and features for both traditional and GenAI models, Amazon Bedrock features, AWS data storage and processing services (S3, Glue, EMR, Kinesis), AWS monitoring and logging tools, and IAM and security best practices. Prior experience with the MLA-C01 exam or foundational AWS certifications (such as AWS Certified Cloud Practitioner or AWS Certified Developer) is beneficial but not required.
Exam Format
The currently bookable MLA-C02 beta exam has 85 multiple-choice and multiple-response questions, a 170-minute time limit, and a $75 USD beta price. AWS explains that beta exams contain additional statistical-evaluation items that do not affect the score. The published standard MLA-C02 exam guide describes 50 scored questions plus 15 unidentified unscored questions, for 65 total, and a minimum passing score of 720 on a 100–1,000 scale. Unanswered questions are incorrect and there is no guessing penalty. AWS uses compensatory scoring, so candidates need to pass the exam overall rather than every domain separately. Beta delivery begins September 29, 2026; AWS has not yet announced general-availability registration or delivery dates. The standard Associate exam price is $150 USD.
Skills Measured
- 1.Content Domain 1: Data Preparation for ML and AI (28% of scored content) — Covers collecting and storing data from sources such as Amazon S3, RDS, DynamoDB, and streaming services like Kinesis and Kafka; feature engineering and transformation using Glue, SageMaker Data Wrangler, and Spark; configuring vector databases for AI applications (OpenSearch Service, RDS with pgvector); preparing data for RAG (chunking, metadata extraction), FM fine-tuning, and continuous pre-training; and validating data quality, managing bias, and ensuring AI training data integrity including prompt-response pair validation.
- 2.Content Domain 2: ML Model and Foundation Model (FM) Development (24% of scored content) — Covers selecting and evaluating FMs from Amazon Bedrock based on task and performance requirements; choosing between custom solutions, managed services, and pre-trained models; training, fine-tuning, and customizing models including task-specific prompt engineering; optimizing retrieval components and embedding models; analyzing model performance using both traditional metrics and GenAI-specific metrics (BLEU, ROUGE, BERTScore, semantic similarity); and implementing LLM-as-a-judge evaluation frameworks and human-in-the-loop workflows.
- 3.Content Domain 3: Deployment and Orchestration of ML and AI Workflows (24% of scored content) — Covers managing deployment infrastructure for traditional ML models and FMs, including Amazon Bedrock Custom Model Import for externally built models; deploying and configuring Bedrock agents and agentic workflow infrastructure; configuring RAG system deployments (retrieval strategies, reranking, knowledge base indexing); implementing CI/CD and automated orchestration pipelines for MLOps and AI workloads; managing prompt versioning via Amazon Bedrock Prompt Management; and configuring GPU resource scaling for AI-specific workloads.
- 4.Content Domain 4: Operating, Monitoring, and Securing ML and AI Solutions (24% of scored content) — Covers monitoring ML and FM model inference performance, including Amazon Bedrock evaluations; detecting and responding to agent coordination failures and tool failures; optimizing infrastructure costs including FM inference costs, token usage, embedding computation, and vector database storage; securing CI/CD pipelines using Amazon CodeGuru and Amazon Inspector; managing access to FMs via Bedrock API keys and IAM credentials; and implementing responsible AI safeguards using Amazon Bedrock Guardrails for sensitive data protection and content safety.
Study Tips
- Start with the official MLA-C02 Exam Guide on AWS Documentation (docs.aws.amazon.com), which lists all four content domains, task statements, and granular skill requirements. Pay particular attention to the new content additions versus MLA-C01 to understand where the exam has shifted toward GenAI.
- Prioritize hands-on experience with Amazon SageMaker AI and Amazon Bedrock. Focus specifically on Bedrock features new to MLA-C02: Bedrock Agents, Bedrock Knowledge Bases, Bedrock Guardrails, Bedrock Prompt Management, and Custom Model Import. Free-tier and lab environments on AWS Skill Builder are available for practice.
- Build working knowledge of RAG architecture end-to-end: data chunking and metadata extraction, embedding model configuration, vector database setup (OpenSearch Service, RDS with pgvector), retrieval pipeline implementation, and reranking. This topic spans all four domains of the exam.
- Study AWS Glue, SageMaker Data Wrangler, SageMaker Feature Store, and AWS Glue DataBrew for the data preparation domain, which carries the highest weight at 28%. Focus on data transformation, bias detection, and preparing multimodal datasets (text, image, audio) for both traditional ML and FM workflows.
- Use the AWS Skill Builder platform, which offers official MLA-C02 exam prep courses, practice question sets, and hands-on labs. The 'Machine Learning Engineer Learning Plan' on Skill Builder is specifically aligned to this exam.
- Review the MLA-C01 to MLA-C02 comparison page on AWS Documentation to understand exactly which skills were added, removed, or recategorized. Many MLA-C01 study materials remain relevant for the 72% of content that overlaps between versions.
- Practice with timed mock exams to build comfort with multiple-response questions, which require selecting all correct answers for full credit. Focus on questions that test cost optimization trade-offs (e.g., FM inference cost vs. latency) and security decision-making (IAM vs. Bedrock API keys, Guardrails configuration), as these appear across multiple domains.
Career Benefits
Earning the AWS Certified Machine Learning Engineer - Associate positions candidates for roles such as ML engineer, MLOps engineer, LLMOps engineer, and data engineer at organizations deploying AI workloads on AWS. According to the World Economic Forum, demand for AI and Machine Learning Specialists is projected to grow by 40%, while 70% of North American IT leaders report difficulty filling AI/ML specialist roles—making certified practitioners especially sought after in the current hiring market.
MLA-C02's expanded coverage of generative AI, Amazon Bedrock, agentic workflows, and RAG architectures differentiates it from older ML certifications and aligns directly with enterprise priorities around deploying production GenAI systems. Professionals holding this certification can demonstrate practical competency in the full ML and GenAI engineering lifecycle—from data ingestion and model development through deployment, monitoring, cost optimization, and security—making it a strong credential for both career entry into ML engineering and advancement into senior MLOps or LLMOps roles.
Sample Questions
5 sample questions with answers and explanations. The full bank has 321 questions, enough for 4 full-length practice exams.
Preview — answers shown1. A custom SageMaker training job must be reproducible after the input data changes. The team also needs to compare artifacts from several runs and promote only the approved one. Which three practices are appropriate? (Select three!)
Multiple correct answersARecord immutable input-data identifiers, code or container version, hyperparameters, and evaluation metrics for each run
BStore the resulting model artifact with a distinct version or registered model package
CUse a model registry approval state before production deployment
DOverwrite the same model artifact path after every run to simplify consumers
EUse the latest training output without retaining its exact data and configuration
FSkip evaluation metadata because the artifact alone determines reproducibility
Explanation
Reproducibility requires preserving the exact data reference, code or image, parameters, and metrics. Distinct artifacts or registered model-package versions preserve lineage and enable comparison, while approval states support controlled promotion. Overwriting paths destroys provenance, “latest” is not reproducible, and an artifact without its inputs and configuration cannot fully explain how it was produced.
2. A company must serve a 250-MB model from Lambda, and the deployment team wants to use a ZIP package with layers. Which constraint determines whether this packaging approach is viable? (Select one!)
AThe compressed ZIP archive must be no more than 50 MB
BThe model must be downloaded from S3 before every invocation
CThe unzipped function package and layers must total no more than 250 MB
DThe function can use at most two layers
Explanation
Lambda ZIP-based deployments have a 250-MB unzipped limit across the function package and layers, and a function can use up to five layers. The compressed archive limit is not the governing constraint described here. Downloading from S3 is an optional design that trades deployment size for runtime latency; it is not required before every invocation. A container image is an alternative with a 10-GB limit.
3. A training dataset has severe class imbalance. The team wants preparation-time methods that can reduce the majority class, increase the minority class, or synthesize minority examples. Which three SageMaker Data Wrangler operators provide these capabilities? (Select three!)
Multiple correct answersARandom undersampling
BRandom oversampling
CSMOTE
DGlue FindMatches
EClarify pre-training DPL
FGlue Schema Registry
Explanation
SageMaker Data Wrangler provides random undersampling to reduce majority examples, random oversampling to increase minority examples, and SMOTE to create synthetic minority examples. FindMatches performs record matching and deduplication, not class balancing. Clarify DPL measures label-distribution bias rather than changing the dataset. Schema Registry enforces streaming schemas and does not rebalance classes.
4. A machine-learning organization needs to select evaluation tools. It will evaluate a foundation model for toxicity, semantic-perturbation robustness, and factual knowledge, and separately needs predictive-model bias detection and explainability. Which pairing is appropriate? (Select one!)
AUse SageMaker Debugger for factual knowledge and Bedrock Guardrails for statistical bias metrics
BUse Clarify for all evaluations because Bedrock Evaluations cannot assess foundation models
CUse fmeval for the foundation-model evaluations and SageMaker Clarify for predictive-model bias and explainability
DUse Model Monitor for toxicity and C3R for predictive-model explainability
Explanation
The fmeval library supports toxicity, robustness through semantic perturbations, and factual-knowledge evaluation for generative models. SageMaker Clarify remains the appropriate tool for predictive-model bias detection and explainability; Bedrock Evaluations does not replace those capabilities. Model Monitor, Debugger, C3R, and Guardrails serve monitoring, debugging, privacy, or safety purposes rather than this complete evaluation pairing.
5. A nonprofit has a small analytics team and receives inconsistent survey files in Amazon S3. They need to interactively inspect samples, clean values, and normalize columns without writing code or managing Spark infrastructure. Which AWS service is most appropriate? (Select one!)
AAWS Glue Data Catalog
BAmazon Managed Service for Apache Flink
CAWS Glue DataBrew
DAmazon SageMaker Feature Store
Explanation
AWS Glue DataBrew provides a visual, point-and-click interface for cleaning and normalizing data without code and includes more than 250 built-in transformations. AWS Glue Data Catalog stores metadata rather than interactively transforming records. SageMaker Feature Store manages reusable ML features, not general-purpose visual preparation. Managed Service for Apache Flink performs stateful stream processing and is not a no-code data-cleaning workspace.
More AWS Practice Exams
AWS Certified DevOps Engineer - Professional (DOP-C02)
DOP-C02 · 436 questions
AWS Certified Generative AI Developer - Professional (AIP-C01)
AIP-C01 · 1978 questions
AWS Certified Machine Learning Engineer - Associate (MLA-C01)
MLA-C01 · 582 questions
AWS Certified Machine Learning - Specialty (MLS-C01)
MLS-C01 · 860 questions
AWS Certified Security - Specialty (SCS-C03)
SCS-C03 · 2069 questions
AWS Certified Solutions Architect - Associate (SAA-C03)
SAA-C03 · 600 questions
$17.99
One-time access to this exam
321 questions (4 practice exams' worth)
Unlimited timed exam simulations
Yours forever — no subscription
Detailed explanations
Free preview stays available