← Back
AWS ML Blog

Announcing AI data preparation capabilities in Amazon SageMaker: Build high-quality datasets for model development workflows

•8 min read•
#amazon#deployment
Level:Advanced
For:AI Engineers
✦TL;DR

Amazon SageMaker now includes AI Data Preparation capabilities that let engineers generate synthetic data, auto‑label datasets, evaluate data quality, and apply transformations via Python SDK APIs and the AWS AI‑ML agent skill, all without building custom pipelines. The auto‑labeling feature uses a jury of multiple models to assign confidence scores, while low‑confidence rows are routed to a human review queue powered by Label Studio Community Edition. Built‑in data quality checks detect duplicates, label noise, and answer leakage, and the synthetic data generator produces code‑based pipelines that can be inspected and scaled. These tools streamline dataset creation and cleaning, accelerating model fine‑tuning and evaluation for production generative‑AI systems.

⚡ Key Takeaways

  • Auto‑labeling supports classification, named entity recognition, extractive question answering, ordinal rating, and summarization, and can operate from an unlabeled set, a small seed set, or no labels.
  • A jury of several models labels each row for fixed‑answer tasks, producing a confidence score used to trigger human review.
  • Human review integrates with SageMaker Studio via the open‑source Label Studio Community Edition, feeding corrections back into the auto‑labeling loop.
  • Synthetic data generation creates a code‑based data generation pipeline grounded in user‑provided seed data and documents, which can be reviewed, modified, and scaled.
  • Data quality evaluation includes built‑in metrics, LLM‑as‑a‑judge metrics, duplicate detection, and checks for label noise or answer leakage.

🔧 Tools & Libraries

SageMaker Python SDKAWS AI‑ML agent skillLabel Studio Community Edition
💡 Why It Matters

By automating labeling, synthetic data creation, and quality checks within SageMaker, engineers can cut weeks of manual data wrangling, reduce labeling costs, and improve model performance while staying compliant with privacy constraints.

✅ Practical Steps

  1. Use the SageMaker Python SDK to call the auto‑labeling API with a labeling spec and seed data to generate labeled datasets.
  2. Route rows with confidence scores below your threshold to the SageMaker Studio review queue powered by Label Studio for expert correction.
  3. Invoke the synthetic data generation API to produce a code‑based pipeline, then edit and scale the generated code as needed for training or evaluation.

Want the full story? Read the original article.

Read on AWS ML Blog ↗

More like this

Introducing workload recovery and expanded infrastructure checks on Amazon SageMaker HyperPod

AWS ML Blog•#deployment

Introducing Unity Gateway Insights

Databricks Blog•#deployment

Introducing Gemini 4 Neon

Google DeepMind•#llm

DaisyRose-27B: Pushing the Frontier of Post-Training Efficiency

Hugging Face Blog•#llm

EXPLORE AI NEWS

Daily hand-picked stories on LLMs, RAG, agents and production AI — curated for engineers who ship.

BROWSE NEWS

GET THE WEEKLY DIGEST

Join engineers getting the Monday signal-over-noise AI breakdown. No spam, unsubscribe anytime.

LEARN AI ENGINEERING

Curated courses, research papers, repos and tutorials built for engineers leveling up in AI.

START LEARNING