Innovation & Tech

AWS Adds Five Data-Preparation Tools to Amazon SageMaker AI

AWS announced five Amazon SageMaker AI data-preparation tools on November 17, 2026, linking synthetic generation, evaluation, labeling and human review.

By Grace Kim

4 min read

AWS Adds AI Data Preparation Tools to Amazon SageMaker AI
AWS Adds AI Data Preparation Tools to Amazon SageMaker AILoboStudioHamburg / pixabay

What's News

  • AWS announced five Amazon SageMaker AI data-preparation capabilities on November 17, 2026.
  • The tools are generally available in two AWS regions: US East (N. Virginia) and US West (Oregon).
  • AWS documentation specifies SDK version 3.30.0 or later and three DataPreparer methods: generate(), evaluate() and transform().
  • Auto-labeling uses model agreement as a confidence score and sends below-threshold rows to human review.
  • Splunk used synthetic data generation while building evaluation models for its Agent Observability product.

AWS announced five Amazon SageMaker AI data-preparation capabilities on November 17, 2026: synthetic data generation, quality evaluation, transformation, auto-labeling and human review. The tools are generally available in the US East (N. Virginia) and US West (Oregon) regions.

AWS said the release consolidates work that often sits across separate labeling tools, generation scripts and validation jobs. The company positioned the update as one managed dataset path from preparation into model development.

What does the new workflow change?

AWS says customers can use one capability inside an existing process or chain several together. Specifications and code remain available for inspection and changes as the dataset moves toward training or evaluation.

The practical gain is fewer handoffs between generation, checking and correction.

How does AWS generate and check data?

AWS's synthetic generator creates training, evaluation and test records from a specification. Customers can optionally ground the output in seed data and documents. AWS also produces a code-based pipeline that customers can review, modify, reuse and scale, making the generation process itself inspectable.

AWS described two ways to check and reshape those records:

  • Quality evaluation applies a customer's business rules, built-in metrics and LLM-as-a-judge metrics, in which a language model assesses the data. It can find duplicates and near-duplicates and surface label noise or answer leakage.
  • Transformation reshapes records through generated code, built-in deduplication, merging, splitting and filtering, or user-supplied Python running at scale.

AWS's example training chain formats the prepared data for supervised fine-tuning, direct preference optimization or reinforcement learning, checks it for quality, and then moves it into SageMaker AI model customization. The sequence places dataset validation ahead of training compute, giving teams a chance to address defective records before a run.

How does auto-labeling reach human review?

AWS's auto-labeling can start with an unlabeled dataset, a small labeled seed set or no labels. Built-in tasks include classification, named entity recognition, extractive question answering, ordinal rating and summarization, and customers can define custom tasks. AWS validates a short labeling specification against the user's examples before processing the full dataset.

For tasks with a fixed set of possible answers, several models label each row as a jury. Their agreement becomes the row's confidence score. AWS sends rows below the selected threshold to a SageMaker Studio review queue based on the open-source Label Studio Community Edition.

Outside subject-matter experts can correct those rows in the review app. Their corrections feed back into auto-labeling, and teams can generate consensus metrics and labeling reports. Human reviewers therefore concentrate on the rows that the model jury cannot settle confidently.

How do managed jobs hand off datasets?

Developers can call these stages through the SageMaker Python SDK or give a coding agent the AWS AI-ML skill to select and run them from SageMaker Studio JupyterLab, Kiro or a local IDE.

AWS documentation specifies SDK version 3.30.0 or later and a DataPreparer class with generate(), evaluate() and transform() methods. The generate() method covers both synthetic data generation and auto-labeling.

Synthetic generation, auto-labeling, quality evaluation and transformation run as SageMaker-managed jobs. Data-producing jobs write records to Amazon S3 and register them in an account hub as DataSet content. For those outputs, the SDK's DataSet object carries an Amazon Resource Name, or ARN, that can be passed to the next stage.

Transformation publishes to the same hub as its input, usually as a new version of that dataset; quality evaluation instead writes results to S3 without registering another dataset. The ARN is the handoff between managed stages.

Where do approvals and customer data sit?

AWS's coding assistant drafts specifications, checks the environment and requests approval before running a job or creating resources. It can also save each job run as a notebook or script, creating a reusable artifact that can run again.

The workflow uses a customer-controlled AWS Identity and Access Management execution role. Source data, generated datasets and labels stay in the customer's own S3 buckets in the selected AWS Region, keeping storage and access within the permissions and location the customer selects. Automation does not displace the customer's control over storage, access or job approval.

What did Splunk achieve with the tools?

Splunk, a Cisco company, used synthetic data generation while building evaluation models for its Agent Observability product. Yash Sheth, director of engineering for Splunk Agent Observability, said the tools produced “statistically representative datasets quickly” while preserving the patterns its models depend on. He also said the SageMaker integration “reduced data wrangling” for data scientists and customers.

The practical test is whether developers can move raw records into validated training data without rebuilding the pipeline at every handoff.

Source: unite.ai

Share this article:

More from Grace Kim

Grace Kim

Show full bio

Market editor covering people, launches and day-to-day practice in entrepreneurship, startups and venture funding at Business Bearings.

2,301 articles

Related articles

« Previous articleNext article »