Why data preparation matters
For teams shipping generative AI to production, data preparation is often the slowest part of the project. Labeled examples are scarce and expensive, privacy rules limit what can be used, and issues such as duplicates, label noise, and answer leakage often go unnoticed until a training run disappoints.
AWS said it is expanding Amazon SageMaker AI with new AI data preparation capabilities aimed at removing the need to stitch together labeling tools, generation scripts, and validation jobs.
Key capabilities
- Auto-labeling: Label data for tasks including classification, named entity recognition, extractive question answering, ordinal rating, and summarization. For tasks with a fixed set of answers, a jury of models labels each row and their agreement becomes the confidence score.
- Human review: Rows below a confidence threshold go to a review queue in SageMaker Studio based on the open-source Label Studio Community Edition. Reviewer corrections feed back into auto-labeling.
- Synthetic data generation: Generate training, evaluation, and test data from a specification, grounded in seed data and documents, with a code-based pipeline that can be reviewed and modified.
- Data quality evaluation: Check datasets against business rules, built-in metrics, and LLM-as-a-judge metrics to surface duplicates, label noise, and answer leakage.
- Data transformation: Describe changes in plain language to generate code, use built-in operations such as dedupe, merge, split, and filter, or run custom Python at scale.
Where it runs
The capabilities can be used directly through the SageMaker Python SDK or invoked by a coding agent through the AI-ML skill in the agent toolkit. AWS said users can work in SageMaker Studio JupyterLab, Kiro, or a local IDE with their coding assistant of choice.



