AWS Certified Machine Learning Engineer – Associate Cheat Sheet (MLA-C01)
Every in-scope SageMaker feature, MLOps workflow, monitoring and security decision the MLA-C01 exam actually tests — organised by the four exam domains with runbooks and decision tables
About This Cheat Sheet
A 14-page associate-tier reference covering every in-scope service from the official MLA-C01 exam guide, organised around the four exam domains: Data Preparation for ML (28%), ML Model Development (26%), Deployment & Orchestration of ML Workflows (22%), ML Solution Monitoring, Maintenance & Security (24%). Structured for candidates with around one year of hands-on SageMaker plus general software engineering — sections move from ML vocabulary and lifecycle (supervised / unsupervised / self-supervised / RL, bias vs variance, splits, cross-validation, feature engineering, common algorithms, neural-network families) through a metrics cheat table (binary / multiclass / regression / ranking / forecast / NLP / clustering) with imbalanced-data guidance, data-preparation services (S3 with Lifecycle + Object Lock + Access Points, Glue + Glue DataBrew + Glue Data Quality DQDL, Lake Formation FGAC, Athena, Redshift + Redshift ML, EMR, Kinesis + Kinesis Video Streams + Data Firehose + Managed Service for Apache Flink, OpenSearch with k-NN + Neural plugin, QuickSight, DataSync + Storage Gateway), SageMaker Data Prep (Data Wrangler, Processing, Ground Truth with Mechanical Turk, Feature Store offline + online, Clarify bias + FM eval), the full SageMaker platform (Studio + Unified Studio + Canvas + JumpStart + Autopilot, training with built-in algorithms + Script Mode + BYO containers + SMDDP + SMP + Managed Spot + Hyperparameter Tuning + Experiments + Debugger/Profiler + HyperPod, inference with Real-time + Serverless + Async + Batch Transform + Multi-Model / Multi-Container + Neo + Production Variants + Inference Recommender, security with execution roles + network isolation + VPC mode), SageMaker MLOps (Pipelines with full step reference, Model Registry approvals, Projects / Service Catalog, Experiments + Lineage Tracking + MLflow managed tracking server, MWAA, Step Functions, EventBridge), model monitoring (Model Monitor four types — Data Quality / Model Quality / Bias Drift / Feature Attribution Drift, Clarify explainability + SHAP, CloudWatch + X-Ray, endpoint Auto Scaling), purpose-built AI services (Bedrock, Comprehend + Comprehend Medical, Rekognition, Textract, Transcribe, Polly, Translate, Lex, Kendra, Personalize, Fraud Detector, Lookout for Equipment / Metrics / Vision, HealthLake, A2I, Q, CodeGuru + DevOps Guru, Amazon Forecast legacy note), compute (EC2 GPU + Trainium + Inferentia + Capacity Blocks, Lambda, Batch, SAR, ECS/EKS/ECR, Auto Scaling, Compute Optimizer), orchestration + application integration (Step Functions, MWAA, EventBridge, SNS + SQS, API Gateway), storage (S3 + EBS + EFS + FSx for Lustre / NetApp / OpenZFS / Windows + S3 Glacier + Storage Gateway), databases supporting ML apps (RDS, DynamoDB, DocumentDB, ElastiCache, Neptune), security (IAM execution roles, KMS, Secrets Manager, Macie, CloudTrail, Config, Organizations with SCPs) + encryption patterns (at-rest + in-transit + VPC mode + network isolation), networking (VPC + gateway/interface endpoints + Direct Connect + CloudFront + API Gateway), observability + cost + governance (CloudWatch + Logs, X-Ray, Chatbot, Systems Manager, Trusted Advisor, Billing + Budgets + Cost Explorer, Service Catalog, CloudFormation + CDK, CodePipeline + CodeBuild + CodeDeploy + CodeArtifact), five reference architectures (end-to-end SageMaker training + deployment, Feature Store-backed real-time inference, multi-tenant Multi-Model Endpoint, streaming anomaly detection, RAG on Bedrock), 25-scenario answer patterns, 12 common pitfalls, a data prep + feature engineering playbook (cleaning, encoding, scaling, text/image/time-series, leakage traps), a SageMaker Pipelines step reference (all 11 step types), four operational runbooks (training OOM, endpoint latency miss, drift detection, runaway cost), and a 23-row quick-pick decision table. Every service on the MLA-C01 in-scope list is audited present.
What's Inside
ML Vocabulary & Lifecycle
Supervised / unsupervised / self-supervised / reinforcement learning, bias-variance tradeoff (with underfit / overfit fixes), training / validation / test splits, cross-validation, feature engineering (scaling, encoding, imputation, binning, interactions, embeddings), common algorithms (Linear / Logistic, Trees, Random Forest, XGBoost, LightGBM, K-Means, DBSCAN, SVM, PCA, RCF, Factorization Machines, DeepAR, BlazingText), neural-network families (MLP, CNN, RNN / LSTM / GRU, Transformer, GAN, Diffusion), and a full metrics cheat table (binary / multiclass / regression / ranking / forecast / NLP / clustering) with imbalanced-data guidance.
Data Preparation
Amazon S3 (Lifecycle + Object Lock + Access Points + SSE-KMS), AWS Glue + Glue DataBrew + Glue Data Quality DQDL, AWS Lake Formation fine-grained access, Amazon Athena, Amazon Redshift + Redshift ML, Amazon EMR + EMR Serverless, streaming (Amazon Kinesis + Kinesis Video Streams + Amazon Data Firehose + Managed Service for Apache Flink), Amazon OpenSearch Service with k-NN + Neural plugin, Amazon QuickSight, AWS DataSync + AWS Storage Gateway. SageMaker-native prep: Data Wrangler visual prep, Processing jobs, Ground Truth labeling (with Mechanical Turk workforce), Feature Store with online + offline stores for point-in-time-correct training, SageMaker Clarify pre-training + post-training bias + SHAP + FM evaluation.
Amazon SageMaker Core
Studio + Unified Studio + Canvas + JumpStart + Autopilot. Training: built-in algorithms (XGBoost, Linear Learner, K-Means, PCA, KNN, RCF, Factorization Machines, IP Insights, BlazingText, Object2Vec, DeepAR, Seq2Seq, Image Classification, Object Detection, Semantic Segmentation, Neural Topic Model, LDA), custom containers + Script mode, distributed training (SMDDP + SMP), Managed Spot + S3 checkpoints, Hyperparameter Tuning (Bayesian / Random / Grid / Hyperband), Experiments, Debugger + Profiler, HyperPod for FM-scale training. Inference: Real-time, Serverless Inference, Asynchronous Inference, Batch Transform, Multi-Model + Multi-Container endpoints, Neo compilation, Production Variants with A/B + shadow, Inference Recommender. Security: execution roles, network isolation, VPC mode, KMS, inter-container encryption.
MLOps — Pipelines & Registry
SageMaker Pipelines with full step reference (ProcessingStep, TrainingStep, TuningStep, ModelStep, TransformStep, ConditionStep, CallbackStep, LambdaStep, EMRStep, ClarifyCheckStep, QualityCheckStep, AutoMLStep, FailStep, NotebookJobStep). SageMaker Model Registry with versioning + approval gates + EventBridge → CodePipeline cross-account deploy. SageMaker Projects with Service Catalog templates. SageMaker Experiments + Lineage Tracking + managed MLflow tracking server. Amazon MWAA for Airflow-native DAGs. AWS Step Functions for heterogeneous durable orchestration. Amazon EventBridge for training-job-state-change + endpoint alarms + registry approvals.
Monitoring, Drift & Explainability
SageMaker Model Monitor four types — Data Quality (missing, types, distributions), Model Quality (accuracy / precision / recall — needs ground-truth capture), Bias Drift, Feature Attribution Drift — with DataCaptureConfig on endpoints and baseline jobs. SageMaker Clarify explainability at inference time (SHAP) + bias reports pre + post training. Amazon CloudWatch default endpoint metrics (Invocations, ModelLatency, OverheadLatency, InvocationsPerInstance, CPU/GPU/MemoryUtilization) plus custom metrics. AWS X-Ray distributed tracing. Endpoint Auto Scaling (target tracking on InvocationsPerInstance or custom metric; scale-to-zero only via Serverless Inference).
Purpose-Built AI Services
Amazon Bedrock managed FMs (Nova + Titan + Claude + Llama + Mistral + Cohere + AI21 + Stability) with Guardrails + Knowledge Bases + Agents + Prompt Management + Model Evaluation. Amazon Comprehend + Comprehend Medical (HIPAA). Rekognition (images + video + custom labels). Textract (forms + tables + lending + invoice). Transcribe (ASR, diarisation, Call Analytics, medical). Polly (TTS, neural voices). Translate (NMT + Active Custom). Amazon Lex (chatbots). Amazon Kendra (enterprise semantic search). Amazon Personalize (recommendations). Amazon Fraud Detector. Amazon Lookout for Equipment / Metrics / Vision. AWS HealthLake (FHIR + NLP). Amazon Augmented AI (A2I) human review. Amazon Q (Business + Developer + in QuickSight). Amazon CodeGuru + DevOps Guru. Amazon Forecast legacy note (exam expects SageMaker Canvas / DeepAR / Chronos for modern forecasting).
Compute, Containers & Storage
EC2 instance families for ML (p5 / p4d / p4de NVIDIA, trn2 / trn1 Trainium, inf2 / inf1 Inferentia, g5 / g6) with Capacity Blocks for ML. AWS Lambda for glue and async preprocessing. AWS Batch for job queues + array jobs. AWS Serverless Application Repository. ECS + Fargate, EKS + Karpenter, Amazon ECR for images with Enhanced scanning. AWS Auto Scaling + AWS Compute Optimizer. Storage: S3 + S3 Glacier, EBS gp3 / io2, EFS for shared notebooks, FSx for Lustre for distributed training scratch (S3 linked), FSx for NetApp ONTAP / OpenZFS / Windows File Server for enterprise shared storage, AWS Storage Gateway for hybrid seeds.
Databases for ML Apps
Amazon RDS (relational source for training data + Redshift ML via SQL), Amazon DynamoDB (online feature serving + session state + inference logging with Streams to Lambda for retraining), Amazon DocumentDB (JSON training data + metadata catalogs with vector search), Amazon ElastiCache (Redis OSS / Valkey / Memcached — semantic cache in front of Bedrock, hot-feature cache, prompt outputs), Amazon Neptune (graph DB for entity features + Neptune Analytics for GraphRAG).
Security, Networking & Governance
IAM execution roles scoped by resource ARN + conditions; permission boundaries for ML teams. KMS CMKs across S3 artefacts / EBS / endpoints / Feature Store offline. AWS Secrets Manager for DB + third-party API rotation. Amazon Macie for PII in training data. AWS CloudTrail org trail + AWS Config managed rules (sagemaker-endpoint-configuration-kms-key-configured, sagemaker-notebook-no-direct-internet-access) + Conformance Packs. AWS Organizations SCPs pinning Region + denying unapproved services. Networking: VPC with private subnets + flow logs, gateway endpoints for S3 / DDB, PrivateLink interface endpoints for SageMaker API + runtime + ECR + STS + CloudWatch + Glue + Athena + KMS + Secrets Manager, AWS Direct Connect for on-prem hybrid, CloudFront + API Gateway + WAF + authorizers for inference fronting.
Observability, Cost & DevOps
CloudWatch metrics + Logs + Logs Insights + subscription filters → Firehose → OpenSearch; AWS X-Ray distributed tracing; AWS Chatbot → Slack / Teams / Chime; AWS Systems Manager (Parameter Store + Patch Manager + Change Manager + Incident Manager); AWS Trusted Advisor security + cost checks; AWS Billing + AWS Budgets with actions + AWS Cost Explorer; AWS Service Catalog for approved SageMaker templates; AWS CloudFormation + AWS CDK for IaC; AWS CodePipeline + CodeBuild + CodeDeploy + CodeArtifact for CI/CD. Cost patterns: Managed Spot training with checkpoints, Serverless / Async inference, Multi-Model / Multi-Container endpoints, Inference Recommender, Graviton / Inferentia / Trainium, FastFile mode, delete unused endpoints.
Reference Architectures
Five full walkthroughs: end-to-end SageMaker training + deployment with Pipelines + Clarify + Condition + Model Registry + EventBridge-triggered CodeDeploy blue/green + Model Monitor; Feature Store-backed real-time inference with online + offline reconciliation; multi-tenant inference on a Multi-Model Endpoint with per-tenant CloudWatch + Auto Scaling; streaming anomaly detection with Kinesis + Managed Flink + SageMaker RCF endpoint + EventBridge alerts + nightly retraining Step Function; RAG assistant using Amazon Bedrock Knowledge Bases with Guardrails and Bedrock Model Evaluation + SageMaker Clarify FM evaluation.
Scenarios, Pitfalls, Runbooks & Decision Table
25 scenario → answer mappings (Serverless inference, Batch Transform, Async for large payloads, Multi-Model endpoints, Model Registry approvals, Model Monitor, Macie PII, Managed Spot training, Feature Store, Lookout for Vision, forecasting, Personalize, Comprehend Medical, A2I review, Ground Truth, HyperPod FM training, Experiments + Lineage, JumpStart + LoRA, Inference Recommender, Clarify SHAP, MWAA vs Pipelines, VPC isolation, Lookout for Metrics, Bedrock + KB). 12 common pitfalls (Managed Spot checkpointing, endpoint cost, File vs Pipe vs FastFile, Feature Store consistency, Model Quality needs labels, drift baselines, IAM execution role scope, VPC-only internet, SageMaker Edge Manager deprecation, Clarify facet column requirement, Inference Recommender model package requirement, Bedrock vs JumpStart). Four runbooks (training OOM, endpoint latency SLO miss, drift detected, runaway cost). 23-row quick-pick decision table.
Why This Cheat Sheet Helps
MLA-C01 is the ML engineering exam — it tests the practical choices: which inference option (Real-time / Serverless / Async / Batch / Multi-Model), which orchestration (Pipelines / MWAA / Step Functions), which drift monitor (Data Quality / Model Quality / Bias / Feature Attribution), when to reach for SageMaker JumpStart vs Bedrock, how to wire Model Registry approvals into CI/CD, how to right-size an endpoint without guessing. This cheat sheet puts those decisions side by side with the exact AWS vocabulary examiners use.
It assumes you already understand ML fundamentals (bias / variance / metrics / feature engineering) and basic Python. Use it in the final weeks to map the exam's surface area, reinforce the SageMaker + AI services matrix, and rehearse the four operational runbooks the exam loves: OOM training, endpoint latency miss, drift detection, and runaway cost.
How to Use It
Skim the whole sheet once to see how the four domains map to sections. Then hammer practice exams — for each wrong answer, find the relevant section and study the tables and decision trees. Pay extra attention to sections 2 (Data Preparation), 3 (SageMaker core), 4 (MLOps), 5 (Monitoring + Clarify), 14 (Scenario → Answer Patterns), 15 (Pitfalls), 16 (Data prep + leakage traps), 17 (Pipelines step reference), 18 (Runbooks), and 19 (Quick-Pick decision table).
In the final week, walk through the four domain weights against your confidence: Data Preparation (28%) → sections 2, 16; ML Model Development (26%) → sections 1, 3, 6; Deployment & Orchestration (22%) → sections 3–4, 7–8, 13; Monitoring, Maintenance & Security (24%) → sections 5, 10–12, 18. Pair with CloudNinja's free MLA-C01 practice exam to surface 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-C01 exam.
How much experience do I need before MLA-C01?
AWS recommends around one year of hands-on SageMaker experience plus general software engineering. You should be comfortable with Python, at least one ML framework (scikit-learn, PyTorch, TensorFlow, or XGBoost), and the basics of model training + hosting on SageMaker. If you do not already hold AI Practitioner (AIF-C01) or have production ML experience on AWS, build that foundation first.
Did AWS retire the old ML Specialty (MLS-C01)?
Yes. AWS retired Machine Learning – Specialty (MLS-C01) and split the coverage into the AI Practitioner (AIF-C01, foundational), Machine Learning Engineer – Associate (MLA-C01) for operational ML + MLOps, and the Generative AI Developer – Professional (AIP-C01) for production GenAI development. MLA-C01 is the current associate-tier ML exam for engineers.
Is this cheat sheet updated for the current MLA-C01 exam?
Yes, it is built directly from the current MLA-C01 exam guide (machine-learning-engineer-associate-01) and audited against the full in-scope services list. It reflects current services and recent changes — SageMaker Unified Studio, SageMaker HyperPod for FM training, managed MLflow tracking server, Inference Recommender, Feature Store online + offline, Model Monitor with Bias + Feature Attribution drift, Amazon Bedrock (Guardrails + Knowledge Bases + Agents + Prompt Management), Amazon Q, SageMaker Edge Manager deprecation guidance toward Greengrass + Neo, and Amazon Forecast as legacy (exam prefers SageMaker Canvas / DeepAR / Chronos).