Human Review

Human review lets your team manually score and label AI outputs. Reviewers work through queues of traces or observations, applying structured scores that feed back into your evaluation pipeline as ground truth.

How it fits in the evaluation pipeline

Live traces (production or experiment)
    ↓
Annotation Queue (filtered set for review)
    ↓
Human Reviewers score each item
    ↓
Scores stored → Golden datasets / model feedback

Use human review to:

• Build ground truth — create labeled datasets from real traffic for fine-tuning or benchmarking.
• Quality assurance — catch systematic failures that automated evaluators miss.
• Model feedback loops — surface low-quality outputs for targeted improvement.