From the source
Amazon SageMaker AI now includes AI data preparation capabilities that let teams generate synthetic examples, assess datasets and route uncertain rows to human reviewers.
- The company says data preparation is often the slowest part of a generative AI project, because labeled examples are scarce and quality problems surface only after a disappointing training run.
- Each capability is a standalone building block, so teams can use one alone or chain them from raw data through to training.
- Auto-labeling covers classification, named entity recognition, extractive question answering, ordinal rating and summarization, and custom tasks can be defined.
- For fixed-answer tasks, a jury of several models labels each row and their agreement becomes that row's confidence score.
- Rows below a confidence threshold go to a review queue in SageMaker Studio built on the open-source Label Studio Community Edition, and reviewer corrections feed back into auto-labeling.
- Splunk, a Cisco company, says the synthetic data generation capability helped it build statistically representative datasets for its Agent Observability evaluation models.






