AWS ML Engineer Associate study plan
AWS ML Engineer Associate · Give each week a clear focus.
8 weeks · About 6 hours per week
Data Preparation for Machine Learning
Map every data-preparation task in Domain 1 to a specific AWS service (S3, Glue, Data Wrangler, Athena, Kinesis) and write a one-line justification for each mapping
Configure an S3 bucket layout for raw, processed, and feature data, then load a sample CSV into it from the console and CLI
Run an AWS Glue crawler over the sample data and verify the Data Catalog table schema it produces
Identify which feature-engineering transforms (scaling, encoding, imputation) belong in Data Wrangler versus a training script
Your checkpoint
A one-page service-to-task mapping table plus a working S3 bucket with a crawled Glue table and a screenshot-free written schema summary
Data Preparation for Machine Learning
Clean and transform the sample dataset with a SageMaker Data Wrangler flow and export the resulting processing job
Detect and document class imbalance, missing values, and outliers using Data Wrangler analysis or Athena queries
Build a SageMaker Processing job that runs a scikit-learn preprocessing script and writes train/test splits back to S3
Explain how SageMaker Feature Store would change this pipeline for online versus offline serving
Your checkpoint
A reproducible preprocessing pipeline (Data Wrangler flow or Processing job script) that outputs versioned train, validation, and test splits in S3
Data Preparation for Machine Learning & ML Model Development
Compare built-in algorithms, script-mode frameworks, and JumpStart pretrained models for the sample problem and justify a choice
Train a baseline model with the SageMaker Python SDK using a managed training job and capture metrics in the training logs
Set up automatic model tuning with a small hyperparameter search space and a defensible objective metric
Diagnose underfitting versus overfitting from the training and validation curves of the baseline
Your checkpoint
A training notebook with a baseline model, one tuning job, and a written comparison of baseline versus tuned metrics
ML Model Development
Evaluate the tuned model with task-appropriate metrics (for example F1, AUC, RMSE) and explain why each metric was chosen
Inspect feature importance or SHAP values with SageMaker Clarify and interpret the top contributors
Register the model in the SageMaker Model Registry with a model card and approval status
Document bias and explainability findings and decide whether the model is fit to proceed
Your checkpoint
A model evaluation report containing metric tables, a Clarify explainability output, and a registered model version with an approval decision
Deployment and Orchestration of ML Workflows
Deploy the registered model to a real-time SageMaker endpoint and invoke it with a sample payload
Compare real-time, serverless, batch transform, and asynchronous inference options against latency, volume, and cost constraints
Write an inference script that handles input preprocessing and output postprocessing inside the container
Configure autoscaling policies for the endpoint and state the target metric and thresholds
Your checkpoint
A deployed endpoint that returns correct predictions on a held-out sample, plus a written inference-option trade-off table
Deployment and Orchestration of ML Workflows & ML Solution Monitoring, Maintenance, and Security
Build a SageMaker Pipelines workflow that chains preprocessing, training, evaluation, and conditional model registration
Add a quality-gate condition step that fails the pipeline when the evaluation metric drops below a threshold
Configure a Model Monitor baseline and schedule for data quality and model quality drift
Set up CloudWatch alarms on endpoint latency, error rate, and invocation count with defined thresholds
Your checkpoint
An end-to-end pipeline execution that runs to completion (or correctly fails a gate) with monitoring schedules and alarms attached
ML Solution Monitoring, Maintenance, and Security
Enforce least-privilege IAM roles for the pipeline, endpoint, and training jobs and verify each role's effective permissions
Encrypt data at rest with KMS and in transit, and enable VPC-only mode for training and endpoint traffic
Enable SageMaker Debugger or clarify logging to detect training anomalies and record findings
Simulate a drift alert and write the retraining trigger and rollback procedure
Your checkpoint
A security and operations runbook covering IAM, encryption, network isolation, drift response, and rollback steps, validated against the deployed solution
Data Preparation for Machine Learning & ML Model Development & Deployment and Orchestration of ML Workflows & ML Solution Monitoring, Maintenance, and Security
Re-derive the domain weightings and allocate remaining study time proportionally to weakest areas
Complete timed practice sets weighted toward Domains 1 and 4 and log every missed question by domain
Re-teach the three weakest concepts using hands-on console or SDK exercises rather than reading
Review all prior deliverables and close any gaps between the documented design and the actual deployed resources
Your checkpoint
A scored practice-exam log with per-domain accuracy and a remediation list where every item is marked resolved or scheduled
Milestones
Week 2: You can explain, without notes, how raw data moves from S3 through Glue and a Processing job into versioned train/test splits
Week 4: A tuned, evaluated model is registered in the Model Registry with metrics and explainability evidence that justify approval
Week 6: One pipeline execution runs preprocessing through conditional registration, and monitoring plus alarms are live on the endpoint
Week 8: Practice-exam accuracy is at or above 80% overall with no domain below 70%, and every remediation item is resolved