SM

Amazon SageMaker Guide

Overview, Access, Workflows, Governance & Training

Zoom

How to use this guide

This guide is organized into four tabs: Overview covering what SageMaker is and who it is for, Access & Setup covering how to get started and what admins configure, Workflow & Components walking through the day-to-day user journey and key building blocks, and Governance & Training covering security, cost controls, and free learning resources.

Content is relevant to data scientists, ML engineers, data engineers, software engineers, analytics professionals, and technical leaders. No prior SageMaker experience is required -- basic familiarity with AWS concepts and ML fundamentals is helpful.

Model training Studio notebooks MLOps pipelines Endpoints & inference Model registry Monitoring & drift Governance & cost
Reminder: SageMaker automates a great deal of the ML lifecycle, but it does not make modeling decisions for you. Always validate model outputs, review results against business requirements, and follow your organization's data handling and approval policies before deploying any model to production.

What is Amazon SageMaker?

Amazon SageMaker is a fully managed AWS service for building, training, and deploying machine learning (ML) models at scale. It provides a web-based development environment (SageMaker Studio), managed compute for training jobs, and multiple options for deploying models to secure, scalable endpoints.

Instead of each team stitching together ad hoc notebooks, containers, and servers, SageMaker offers a standardized platform with integrated data preparation, experimentation, MLOps, and monitoring -- making it easier for organizations to move from experiments on laptops to repeatable, production-grade ML workflows.

Key benefits

  • Fully managed infrastructure for training and inference, with on-demand and spot instances
  • A unified environment for notebooks, data prep, training jobs, and deployments
  • Built-in algorithms and support for popular frameworks: TensorFlow, PyTorch, and XGBoost
  • MLOps features including pipelines, a model registry, and monitoring for drift and quality
  • Consistent experience across teams that reduces ramp-up time and infrastructure overhead

Who this training is for

RolePrimary use
Data scientists & ML engineersBuild, train, evaluate, and deploy models
Data engineersPrepare data and manage pipelines into and out of SageMaker
Software engineersIntegrate deployed model endpoints into applications and services
Analytics professionalsTurn insights into reusable, production-ready models
Technical leadersUnderstand platform capabilities, guardrails, and governance levers
Platform / cloud engineersConfigure the SageMaker domain, networking, security, and permissions

What you will be able to do

After completing a standard SageMaker enablement path, users should be able to:

🚀
Launch SageMaker Studio and create a new project or notebook
🔗
Connect to approved data sources and perform exploratory analysis and feature engineering
🧪
Train, evaluate, and compare models using managed training jobs
🌐
Deploy models to real-time and batch endpoints and integrate them into applications
📊
Monitor model performance and data drift, and update models as needed
🔒
Configure the shared SageMaker domain, security, networking, and permissions (platform/cloud engineers)

How SageMaker fits into your AI ecosystem

SageMaker is one piece of a broader data, analytics, and AI stack:

  • Upstream: Data lakes, warehouses, and streaming platforms feed data into SageMaker for feature creation and training
  • Downstream: Applications, APIs, and dashboards consume predictions from SageMaker endpoints
  • Alongside: Code assistants, chat interfaces, and analytics tools can call SageMaker models as part of larger workflows
  • CI/CD: SageMaker Pipelines integrates with existing DevOps tooling for automated model deployment

Access and prerequisites

To use SageMaker, the following must be in place before you can open Studio or run a training job.

User requirements

  • An active AWS account in an approved Region
  • IAM permissions for Amazon SageMaker and Amazon S3 (typically granted via group-based policies)
  • Network connectivity to the configured VPC and endpoints if the organization uses private networking
  • A SageMaker domain created and configured by a platform or cloud team

Most organizations manage SageMaker access via identity providers and group-based permissions. Users are onboarded through existing access request processes rather than direct sign-ups.

Opening SageMaker Studio

  1. Sign in to the AWS Console with your organization credentials
  2. Navigate to Amazon SageMaker
  3. Click Open Studio (or use the organization-provided direct URL)
  4. Select the correct profile if multiple user profiles are available in your domain
First time? Studio may take a minute or two to launch while your user environment provisions. This only happens once -- subsequent launches are faster.

One-time platform setup (admins)

Platform and cloud teams perform a one-time setup to create the shared SageMaker domain and baseline configuration. End users do not need to complete these steps.

1

Create the SageMaker domain

In the AWS Console, open SageMaker and start the "Set up for organizations" flow. Define the Region, domain name, and authentication method.

2

Configure execution roles

Define IAM execution roles that control what SageMaker workloads can access -- S3 buckets, KMS keys, logs, and other services.

3

Set up S3 storage

Select or create S3 buckets for datasets, model artifacts, and logs. Apply bucket policies and encryption settings.

4

Enable applications

Choose which SageMaker applications to enable: Studio, Canvas, Code Editor, and others based on team needs.

5

Configure networking

Define VPC, subnets, and security groups. Choose whether access is internet-facing or fully private with VPC endpoints.

6

Apply org policies and onboard users

Apply encryption, logging, and resource tagging policies. Assign users to groups or profiles with the correct permissions.

Day-to-day user workflow

From an individual user's point of view, SageMaker provides a repeatable path from idea to deployed model. The steps below represent the standard journey -- not every project will use every step.

1

Open SageMaker Studio

Use the organization-provided URL or the AWS Console. Select the right profile if multiple profiles are available in your domain.

2

Start a project or notebook

Create a new project from a template or start with a blank notebook. Connect to data in S3 or other approved sources. Use projects to group related experiments, pipelines, and models.

3

Explore and prepare data

Use notebooks or SageMaker Data Wrangler to clean, transform, and engineer features. Document your data choices so the pipeline can be reproduced.

4

Train and tune models

Launch training jobs using built-in algorithms or custom code. Use hyperparameter tuning jobs to automate the search for better configurations. Track experiments with SageMaker Experiments.

5

Evaluate and register models

Analyze metrics and validation results in Studio. Register selected models in the model registry with versioning and metadata for traceability and approvals.

6

Deploy and monitor

Deploy models as real-time endpoints, serverless endpoints, or batch transform jobs. Configure monitoring to track performance, data drift, and operational health. Iterate as results and data change.

Core SageMaker components

The training program focuses on practical use of these main SageMaker building blocks.

St
SageMaker Studio
  • Web-based IDE for the full ML lifecycle
  • Integrated notebooks, terminals, and file browser
  • Centralized view of experiments, models, and pipelines
  • Support for JupyterLab and Code Editor (VS Code-based)
  • Serverless notebooks (no instance management required)
DW
Data Wrangler
  • Visual data preparation and feature engineering
  • 300+ built-in transforms with no-code interface
  • Data quality and insights reports
  • Connects to S3, Athena, Redshift, and other sources
  • Export flows to pipelines or notebooks
Tr
Training Jobs
  • Managed compute for model training at any scale
  • Built-in algorithms and custom container support
  • Automatic model tuning (hyperparameter optimization)
  • Spot instance training to reduce cost
  • Distributed training for large datasets and models
Rg
Model Registry
  • Centralized catalog of trained model versions
  • Metadata, metrics, and lineage for each version
  • Approval workflows before production deployment
  • Integration with Pipelines for automated registration
  • Supports cross-account and cross-region sharing
Pi
Pipelines
  • Automated, reproducible ML workflows
  • DAG-based step definitions (processing, training, evaluation, registration)
  • Native integration with CI/CD systems
  • Conditional logic and parameter passing between steps
  • Audit trail of every pipeline run
En
Endpoints & Inference
  • Real-time endpoints for low-latency predictions
  • Serverless endpoints for intermittent traffic
  • Batch transform for large-scale offline scoring
  • Multi-model endpoints to reduce hosting cost
  • Auto-scaling based on traffic patterns
Mo
Model Monitor
  • Continuous monitoring of production endpoint quality
  • Data quality checks against a baseline
  • Model quality tracking against ground truth labels
  • Bias drift and feature attribution drift detection
  • Alerts via CloudWatch when violations are detected
Ca
SageMaker Canvas
  • No-code ML for business analysts
  • Point-and-click model building with AutoML
  • Built-in ready-to-use models (forecasting, NLP, vision)
  • Collaboration with data science teams
  • Direct deployment to Studio or production endpoints

Security and governance

SageMaker provides built-in hooks for governance, security, and cost controls that organizations can align with existing cloud standards.

Access and identity

  • IAM roles and policies restrict which data and services SageMaker workloads can access
  • SageMaker domain user profiles control per-user permissions and app access
  • Integration with AWS SSO / IAM Identity Center for federated access

Network security

  • Run SageMaker resources in private subnets with VPC endpoints for data access
  • Disable internet access for Studio notebooks in regulated environments
  • Security groups control traffic to and from SageMaker resources

Data protection

  • Enforce encryption at rest (SSE-KMS) for all S3 data, model artifacts, and volumes
  • Enforce encryption in transit (TLS) for all API calls and data movement
  • Apply S3 bucket policies and S3 Object Lock for sensitive training datasets

Cost controls

  • Apply mandatory resource tags for cost allocation and reporting
  • Use AWS Budgets and alerts to catch runaway training jobs early
  • Use spot instances for training to reduce compute cost by up to 90%
  • Enable lifecycle configurations to auto-shut-down idle Studio kernels and apps
  • Choose the smallest instance type that meets training time requirements
User responsibility: Always shut down notebook kernels and Studio apps when not in use. Idle compute instances continue to accrue costs. When in doubt, check with your cloud team on approved instance types and budget thresholds before starting a large training job.

Sensitive data handling

Training content highlights how users are expected to operate safely:

  • Use only approved data sources -- do not load sensitive data from outside the designated S3 buckets
  • Do not include PII, PHI, or other regulated data in notebook outputs or experiment logs unless explicitly approved
  • Follow your organization's data classification policy when selecting training datasets
  • Use SageMaker's built-in bias and explainability tools (Clarify) to detect and document fairness issues before deploying models

Typical learning timeline

A common enablement pattern moves learners from orientation to first production use in phases.

Phase 1
Intro

SageMaker concepts, Studio interface, and org-specific policies. No hands-on required.

Phase 2
Foundations

Data access, notebooks, training jobs, and simple deployments. Hands-on labs.

Phase 3
Intermediate

Pipelines, model registry, monitoring, and cost optimization strategies.

Phase 4
Advanced

Custom containers, large-scale training, and complex integration patterns.

Foundations -- Free Training

Free official resources from AWS and reputable platforms to get started with Amazon SageMaker.

AWS -- Official Interactive -- Get Started Amazon SageMaker Getting Started Official AWS getting started experience. Step-by-step walkthroughs for launching Studio, running your first notebook, and training a model. Open on AWS --> AWS Skill Builder -- Official Course -- Free -- Self-paced Getting Started with Amazon SageMaker Studio Free AWS-authored course covering Studio setup, notebooks, experiments, and the core ML workflow. Includes knowledge checks and a completion badge. Start on Skill Builder --> AWS -- YouTube Video Playlist -- Beginner Amazon SageMaker -- Official Video Series AWS official YouTube playlist with short explainer videos for each SageMaker feature. Great for visual learners who want a quick introduction to specific components. Watch on YouTube --> AWS Documentation Reference -- Always Current Amazon SageMaker Developer Guide The authoritative reference for all SageMaker features, APIs, and configuration options. Use this alongside training to look up specific details. Open Developer Guide -->
Intermediate to Advanced -- Free Training

Deeper dives into MLOps, pipelines, model monitoring, and large-scale training patterns.

AWS Skill Builder -- Official Learning Plan -- Multi-course AWS Machine Learning Learning Plan Official multi-course plan covering ML fundamentals through advanced SageMaker patterns. Structured path to the AWS Machine Learning Specialty certification. Start on Skill Builder --> AWS -- GitHub Notebooks -- Hands-on Examples Amazon SageMaker Example Notebooks Official AWS GitHub repository with hundreds of working Jupyter notebooks covering every SageMaker feature. The most practical learning resource available. Open on GitHub --> Coursera -- AWS & DeepLearning.AI Specialization -- Free to audit Practical Data Science on AWS (3-course Specialization) AWS and DeepLearning.AI co-authored specialization. Covers SageMaker for data science workflows, NLP, and ML pipelines. Free to audit all three courses. View on Coursera --> AWS Skill Builder -- Official Course -- Free -- Intermediate Amazon SageMaker Studio for Data Scientists Deeper course focused on the data scientist workflow in Studio -- experiments, pipelines, model registry, and deployment. Includes hands-on labs. Start on Skill Builder -->

Feature availability, pricing, and service details may change as AWS updates Amazon SageMaker. Always verify current information at docs.aws.amazon.com/sagemaker and follow your organization's cloud governance and data handling policies.

This guide is for training and orientation only. Follow your organization's data security and acceptable use policies for all SageMaker workloads.