The lifecycle, station by station
Lesson 1 of 4 in The ML Lifecycle on AWS.
Strip away the service names and every ML system on AWS runs the same route: collect data, prepare it, train a model, evaluate it, deploy it, monitor it — then let what monitoring finds start the next lap. The exam guide organizes its four domains along exactly this track, which makes the ML lifecycle the single most useful mental model to carry into the exam room: before reading any option, place the question on the line.
Each station has owners. Amazon S3 anchors collect (with Kinesis and Glue feeding it), the Glue family and SageMaker Data Wrangler own prepare, SageMaker AI training jobs own train, evaluation runs on metrics plus managed MLflow tracking, endpoints and Amazon Bedrock own deploy, and CloudWatch with SageMaker Model Monitor owns monitor.
| Station | Primary owners | Supporting cast | Exam domain |
|---|---|---|---|
Collect | Amazon S3 | Kinesis, Amazon Data Firehose, Glue connectors | Domain 1 (28%) |
Prepare | AWS Glue, SageMaker Data Wrangler | EMR, DataBrew, Feature Store | Domain 1 (28%) |
Train | SageMaker AI training jobs | Automatic model tuning, Spot capacity | Domain 2 (24%) |
Evaluate | SageMaker evaluation + Clarify | Managed MLflow, Bedrock evaluations | Domain 2 (24%) |
Deploy | SageMaker endpoints, Amazon Bedrock | Model Registry, the Code* CI/CD family | Domain 3 (24%) |
Monitor | CloudWatch, SageMaker Model Monitor | CloudTrail, budgets and alarms | Domain 4 (24%) |
Interactive checkpoint quiz (2 questions) — open this page in a browser to take it.