AWS Certified Machine Learning Engineer – Associate Cheat Sheet (MLA-C02)
Every in-scope service and decision the MLA-C02 exam tests — classic SageMaker AI engineering plus Bedrock, RAG, agents and generative AI evaluation, organised by the four exam domains with closed services tagged
Updated for MLA-C02. MLA-C01 retired in English on September 28, 2026.
See the full timelineAbout This Cheat Sheet
- 13 pages
- Every in-scope MLA-C02 service
- Organised by the four MLA-C02 domains
- Bedrock, RAG, agents and GenAI evaluation
- Closed SageMaker AI features tagged
- Updated November 2026
Exam domains
Share of the scored MLA-C02 content
- Data Preparation for ML and AI28%
- ML Model and Foundation Model (FM) Development24%
- Deployment and Orchestration of ML and AI Workflows24%
- Operating, Monitoring, and Securing ML and AI Solutions24%
Services covered
Generative AI
- Bedrock
- Bedrock AgentCore
- Knowledge Bases
- Guardrails
- Prompt Management
- Model evaluation
- Custom Model Import
- Model Distillation
SageMaker AI
- Data Wrangler
- Processing
- Feature Store
- JumpStart
- Canvas
- HyperPod
- MLflow
- Pipelines
- Model Registry
- Inference Recommender
Data preparation services
- S3
- Glue
- Glue DataBrew
- Glue Data Quality
- Lake Formation
- Athena
- Redshift
- EMR
- Kinesis
- Kinesis Video Streams
- Data Firehose
- Managed Service for Apache Flink
- OpenSearch Service
- S3 Vectors
- Amazon Quick
- DataSync
- Storage Gateway
AI services
- Comprehend
- Comprehend Medical
- Rekognition
- Textract
- Transcribe
- Translate
- Polly
- Lex
- Personalize
- HealthLake
Show 58 more servicesShow fewer
Compute & containers
- EC2
- Lambda
- Batch
- ECR
- ECS
- EKS
- Auto Scaling
- Compute Optimizer
Orchestration & integration
- Step Functions
- MWAA
- EventBridge
- SNS
- SQS
- API Gateway
Storage
- EBS
- EFS
- FSx
- S3 Glacier
Databases & vector stores
- RDS
- DynamoDB
- DocumentDB
- ElastiCache
- Neptune
Security
- IAM
- Bedrock API keys
- KMS
- Secrets Manager
- Macie
- Inspector
- CloudTrail
- Config
- Organizations
Networking
- VPC
- Direct Connect
- CloudFront
Observability & cost
- CloudWatch
- X-Ray
- Systems Manager
- Trusted Advisor
- Billing and Cost Management
- Budgets
- Cost Explorer
IaC & CI/CD
- CloudFormation
- CDK
- Service Catalog
- CodePipeline
- CodeBuild
- CodeDeploy
- CodeArtifact
- CodeCommit
- CodeConnections
Closed to new customers
- Model Monitor
- Clarify
- Debugger
- Ground Truth
- Bedrock Agents Classic
- CodeGuru Reviewer
- DevOps Guru
Feature details
- Knowledge Bases
- Chunking strategies · Hybrid search · Reranking · Incremental sync
- AgentCore
- Runtime · Gateway · Memory · Identity · Observability · Evaluations
- Bedrock inference
- On-demand tiers · Batch · Cross-Region · Prompt caching · Inference profiles
- Inference
- Real-time · Inference components · Serverless · Async · Batch Transform · Shadow tests
- Training
- AMT · Distributed training · Managed Spot
Concepts covered
- Foundation models
- Customization ladder: prompting, RAG, fine-tuning, distillation · Embeddings and vector stores · LLM-as-a-judge and RAG evaluation
- Metrics
- Classification, regression, ranking · ROUGE, BLEU, BERTScore
- Exam practice
- Scenario → answer table · Exam tips for every domain
Built from the official MLA-C02 exam guide; every service on its in-scope list is audited present.
What's Inside
ML and Generative AI Vocabulary
The fundamentals the exam assumes, from bias and variance to tokens, embeddings, RAG and agents, plus a metrics table.
- Learning types, bias vs variance, catastrophic forgetting
- Splits: stratified, chronological, group-aware
- Tokens, context window, temperature
- Embeddings and RAG; what an agent is
- Metrics: classification, regression, ranking, retrieval
- Text generation: ROUGE, BLEU, METEOR, BERTScore
- LLM-as-a-judge and RAG retrieval metrics
Domain 1: Data Preparation for ML and AI
Ingesting, cleaning and validating data, and preparing it for foundation models: embeddings, chunking, PII and vector stores.
- S3, Kinesis, Data Firehose, Flink, Glue, DataBrew, Glue Data Quality
- Lake Formation, Athena, EMR, Redshift, Amazon Quick
- Data Wrangler, Processing, Feature Store, training input modes
- Cleaning, encoding, imbalance, leakage, pre-training bias metrics
- Embeddings models and multimodal embeddings
- Chunking: fixed-size, hierarchical, semantic, none
- Redact PII before indexing; fine-tuning data in JSON Lines
- Vector stores: OpenSearch, S3 Vectors, pgvector, Neptune
Domain 2: ML Model and FM Development
Choosing between AI services, Bedrock, JumpStart and built-in algorithms, customising FMs, training on SageMaker AI and evaluating models.
- AI services: Textract, Comprehend, Rekognition, Transcribe, Personalize
- Bedrock vs JumpStart vs built-in algorithms
- Prompting, RAG, fine-tuning, reinforcement fine-tuning, distillation
- Hyperparameters, automatic model tuning, distributed training
- Managed Spot, HyperPod, MLflow on SageMaker AI
- Debugging convergence: exploding and vanishing gradients
- Bedrock evaluations: automatic, LLM-as-a-judge, human
- RAG evaluation, SHAP explanations, shadow tests
Domain 3: Deployment and Orchestration
Hosting models and agents, configuring RAG, and automating the ML and AI lifecycle with pipelines and CI/CD.
- Real-time, inference components, serverless, async, batch
- Bedrock on-demand tiers, batch, cross-Region, Provisioned Throughput
- Custom Model Import for models fine-tuned elsewhere
- Knowledge Bases: sync, hybrid search, reranking, metadata filters
- AgentCore Runtime, Gateway, Memory, Identity, Observability
- EC2 GPU, Trainium, Inferentia; ECS, EKS, Lambda, Batch
- Pipelines, Model Registry, Prompt Management versions
- Step Functions, MWAA, EventBridge, CodePipeline
Domain 4: Operating, Monitoring and Securing
Monitoring models, foundation models and agents, controlling token and infrastructure cost, and securing ML and AI workloads.
- Drift detection pipelines on endpoint data capture
- Bedrock metrics, invocation logging, generative AI observability
- AgentCore Observability traces for agent steps
- Prompt caching, batch inference, application inference profiles
- IAM, Bedrock API keys, KMS, Secrets Manager, Macie
- Guardrails and the ApplyGuardrail API
- Inspector image scanning in CI/CD
- VPC, network isolation, PrivateLink endpoints
Scenarios and Closed Services
A scenario → answer table for quick revision, and the services and features that are closed, retired or out of scope.
- 26 scenario → answer rows across all four domains
- Model Monitor, Clarify, Debugger, Ground Truth: closed
- Bedrock Agents Classic, CodeGuru Reviewer, DevOps Guru: closed
- Topics removed from the MLA-C02 exam guide
- Service renames to recognise on the exam
Why This Cheat Sheet Helps
MLA-C02 keeps the classic ML engineering exam and adds foundation models throughout. You still choose between real-time, serverless, asynchronous and batch inference, pipelines and orchestration, and Spot or on-demand training. Now you also pick a vector store, a chunking strategy, RAG or fine-tuning or distillation, where to host an agent, and how to evaluate, guard and pay for a foundation model. This cheat sheet puts those decisions side by side, in the vocabulary of the MLA-C02 exam guide.
It also flags what changed under the exam: SageMaker Model Monitor, Clarify, Debugger and Ground Truth are closed to new customers, and the MLA-C02 guide no longer names them, so the sheet shows the current way to do each job (drift detection pipelines, MLflow, Bedrock evaluations).
How to Use It
Skim the whole sheet once to see how the four domains map to sections 2 to 5. Then take practice exams and, for each wrong answer, find the matching section and study its tables. The Exam Tip boxes hold the trigger phrase → answer rules that come up most often, and section 7 collects them as a scenario table.
In the final week, check yourself against the domain weights: Data Preparation for ML and AI (28%) → section 2; ML Model and FM Development (24%) → sections 1 and 3; Deployment and Orchestration (24%) → section 4; Operating, Monitoring and Securing (24%) → section 5. Pair it with CloudNinja's free MLA-C02 practice exam to find gaps.
Frequently Asked Questions
Is this AWS Machine Learning Engineer Associate cheat sheet free?
Yes, completely free with no signup required. Download the PDF directly from CloudNinja and use it as a study reference for the MLA-C02 exam.
How much experience do I need before MLA-C02?
AWS recommends at least one year of experience using SageMaker AI, Amazon Bedrock and other AWS services for ML engineering, plus a year in a related role such as backend developer, DevOps engineer, data engineer or data scientist. You should know both traditional ML and generative AI.
What changed from MLA-C01 in this cheat sheet?
It was rebuilt for MLA-C02. It adds foundation model selection and customization, embeddings and vector stores, RAG with Bedrock Knowledge Bases, agents on AgentCore, Prompt Management, generative AI evaluation, Guardrails, Bedrock API keys and token cost. It drops SageMaker Neo, bring-your-own-container and model-size reduction, which the MLA-C02 guide removed.
Is this cheat sheet updated for the current MLA-C02 exam?
Yes. It is built from the MLA-C02 exam guide (machine-learning-engineer-associate-02) and checked against its full in-scope services list. It tags services and features that AWS has closed to new customers, including SageMaker Model Monitor, Clarify, Debugger and Ground Truth, Bedrock Agents Classic, CodeGuru Reviewer and DevOps Guru, and shows what to use instead.
Keep Studying
Recommended courses
Affiliate links — if you enrol through them, CloudNinja may earn a commission at no extra cost to you.