Amazon SageMaker
By Amazon Web Services
Amazon SageMaker is a fully managed AWS platform that covers the end-to-end machine learning lifecycle — data preparation, model building, training, tuning, deployment, and monitoring — without requiring teams to manage the underlying…
Definition
Amazon SageMaker is a fully managed AWS platform that covers the end-to-end machine learning lifecycle — data preparation, model building, training, tuning, deployment, and monitoring — without requiring teams to manage the underlying infrastructure.
Overview
Before platforms like SageMaker, teams building ML models had to stitch together separate tools and infrastructure for each stage of the workflow: notebooks for experimentation, clusters for training, and custom deployment pipelines for serving predictions. SageMaker consolidates these stages into one managed platform — SageMaker Studio provides a notebook environment for development, managed training jobs provision the compute needed to train models built with frameworks like PyTorch or TensorFlow, and built-in hyperparameter tuning automates the search for better model configurations. Once a model is trained, SageMaker handles deployment to scalable, managed endpoints for real-time or batch inference, along with monitoring for data and model drift over time. It also includes tools for data labeling and feature management, and increasingly integrates with generative AI workflows alongside services like Amazon Bedrock for teams that need both custom model training and access to foundation models. Because it spans the full ML lifecycle rather than a single stage, SageMaker is generally positioned for teams doing custom model development, as distinct from Bedrock's focus on consuming and lightly customizing pre-built foundation models — concepts covered in courses like PyTorch Deep Learning and TensorFlow & Keras.
Key Features
- SageMaker Studio notebook environment for ML development
- Managed, scalable training jobs for custom model training
- Automated hyperparameter tuning
- Managed real-time and batch inference endpoints for deployment
- Model monitoring for data and prediction drift
- Built-in data labeling and feature store tooling
- Support for popular ML frameworks including PyTorch and TensorFlow
- MLOps pipelines for reproducible, automated training and deployment