AWS adds data labeling and synthetic-data tools to SageMaker AI

Amazon Web Services added AI data-preparation capabilities to SageMaker AI on November 17, combining automated labeling, human review, synthetic data generation, quality checks and transformation tools for model-development datasets.
The capabilities are generally available in Northern Virginia and Oregon. Developers can invoke them through the SageMaker Python SDK or use the AWS AI-ML agent skill from environments including SageMaker Studio JupyterLab, Kiro and local coding tools.
Automatic labeling can begin with unlabeled records or a small labeled sample. For certain tasks, several models label each record and their agreement determines a confidence score. Records below a chosen threshold can be sent to human reviewers, whose corrections feed back into labeling.
Synthetic-data generation produces a code-based pipeline that developers can inspect and modify. Quality checks look for issues including duplicates, unreliable labels and leaked answers. The workflow runs under a customer-controlled IAM role, with source data, generated datasets and labels kept in the customer’s S3 buckets.
The update puts more of the preparation process alongside model training, aiming to reduce separate pipeline-building work while preserving human review and inspectable code.
Sources and citations
Revision history
Published .