Amazon SageMaker
The AWS workshop for building, training and serving machine learning models.
Amazon SageMaker is the AWS platform for the whole machine learning lifecycle: preparing data, writing and running notebooks, training a model on GPU machines, and deploying it as an endpoint your app can call. Each step is a managed piece. You say 'train this script on four GPU instances', SageMaker launches them, runs the job, saves the model to Amazon S3 and shuts the machines down.
The difference with Amazon Bedrock is who makes the model. Bedrock gives you ready-made foundation models (Claude, Llama, Amazon's own) through an API. SageMaker is for when you train or fine-tune your own, whether a small fraud detector on tabular data or a custom vision model. SageMaker JumpStart sits in between: a catalog of open models you can deploy or fine-tune in a few clicks.
Example: a retailer wants to predict which customers will stop buying. A data scientist explores the data in a SageMaker notebook, launches a training job with XGBoost on past purchases, compares a few experiments, and deploys the best model to a real-time endpoint. Each night a batch job scores every customer and writes the result to a table that marketing reads.
SageMaker has grown into a large family (Studio, Pipelines, Feature Store, Model Monitor, Ground Truth for labeling). Nobody uses all of it; most teams start with notebooks, training jobs and endpoints.
Keep in mind
- Covers data prep, notebooks, training, deployment and monitoring in one managed service.
- Training jobs spin up GPU instances only while they run; endpoints serve predictions over HTTPS.
- Bedrock is for using existing foundation models; SageMaker is for training or fine-tuning your own.
- JumpStart offers open models ready to deploy or fine-tune.
