AWS has introduced AI data preparation capabilities in Amazon SageMaker AI to help data scientists and ML engineers build high-quality training and evaluation datasets. The update brings five capabilities: synthetic data generation, automated labeling, human-in-the-loop review, data quality evaluation, and data transformation.
Teams can generate task-specific examples from their documents, label large datasets with confidence scores, and send uncertain results for expert review.
They can also identify duplicates, detect quality issues, and reshape datasets for supervised fine-tuning, direct preference optimization, and reinforcement learning. Developers can access these capabilities through the SageMaker Python SDK or AI coding assistants.

