Machine Learning
Amazon SageMaker Ground Truth
Amazon SageMaker Ground Truth manages data labeling workflows using private, vendor, or public workforces. It supports human annotations for training and evaluating GenAI systems with task-specific instructions and quality controls.
Key points
- Labeling jobs define input manifests, task UI, workforce, and output location
- Annotation consolidation combines worker responses
- Automated data labeling can reduce human effort for supported tasks
When to use it
- Create a human-rated prompt-response evaluation set
- Label document relevance for a RAG benchmark
Exam tips
- Choose Ground Truth for dataset labeling and A2I for per-prediction human review in a live workflow
- Pilot instructions and measure inter-annotator agreement before scaling a subjective rubric