Amazon SageMaker AI
The end-to-end platform for building, training, and deploying ML models.
❓ What is it?
AWS's managed machine-learning platform covering the whole lifecycle: notebooks (Studio), data prep, distributed training, hyperparameter tuning, model registry, real-time/batch/serverless inference endpoints, and MLOps pipelines.
💡 Why does it exist?
Assembling an ML stack from raw EC2, storage, and schedulers is months of platform work. SageMaker packages the lifecycle so data scientists ship models while AWS handles provisioning, scaling, and patching.
⏱️ When should you use it?
Reach for SageMaker when you are training or hosting your OWN models — custom fraud scoring, forecasting, computer vision. If a prebuilt AI service (Rekognition, Comprehend) or a foundation model API (Bedrock) already solves it, use those first.
🗺️ Where does it fit?
The hub of a custom-ML architecture: it reads training data from S3, runs training jobs on managed instances, registers artifacts, and exposes endpoints that applications call through the SageMaker runtime.
🔌 How do you integrate it?
Work in SageMaker Studio; launch training with built-in algorithms or your own containers; deploy with one SDK call to a real-time, serverless, or asynchronous endpoint; automate the flow with SageMaker Pipelines.
🧩 Commonly integrated with
🎯 Exam angle (AIF-C01)
- Map inference options: real-time (steady traffic), serverless (spiky/intermittent), asynchronous (large payloads, near-real-time), batch transform (whole datasets, no persistent endpoint).
- The classic decision question: prebuilt AI service < Bedrock < SageMaker, ordered by how much ML expertise and control the scenario demands.