Proof of Quality

Sapien's approach to reviewing data against customer criteria, with expert judgments used to evaluate AI judges.

The customer defines what the data needs to pass. Human experts apply those criteria to reference examples, and agents are evaluated on their alignment with those judgments. Review results should distinguish the data's quality from agreement among reviewers. The dataset, thresholds and review workflow are scoped for each pilot.

In practice

A pilot can assess whether task labels match the supplied video evidence before the data enters training.