Amazon has expanded SageMaker AI with new data preparation capabilities aimed at model development workflows such as fine-tuning, reinforcement learning and building evaluation suites. Each capability is a standalone building block that can be used alone or chained together. Auto-labeling labels data from an unlabeled set, a small labeled seed set or no labels, with built-in task types including classification, named entity recognition, extractive question answering, ordinal rating and summarization, plus custom tasks. For fixed-answer tasks, a jury of several models labels each row and their agreement becomes the confidence score.
Rows below a confidence threshold go to a review queue in SageMaker Studio that uses the open-source Label Studio Community Edition, where outside subject matter experts can take part. Other capabilities cover synthetic data generation from a specification grounded in seed data and documents, data quality evaluation against custom rules, built-in metrics and LLM-as-a-judge metrics, and data transformation that can be described in plain language or run with built-in operations such as dedupe, merge, split and filter.
Work can happen in the SageMaker Python SDK or through a coding agent that has the AI-ML skill, in SageMaker Studio JupyterLab, Kiro or a local IDE. Data, generated datasets and labels stay in the customer's own Amazon S3 buckets, and the service runs under an IAM execution role the customer controls.
Why it matters if you build agents
Developers building agents that rely on custom models can generate, label and check training and evaluation data in one workflow instead of wiring together separate labeling, generation and validation jobs.
The SDK takes specifications as input, and the post recommends creating them with the AI-ML skill first, so teams should plan to review the generated specification and code before running large jobs.
Availability is limited to the US East (N. Virginia) and US West (Oregon) Regions for now.
Get stories like this every weekday
One short email with the day's agent tooling news.