Atlas Auditor - Data Annotation Quality Assurance
As an Atlas Auditor, I conducted final audits of motion study video annotations to ensure high accuracy. I resolved disputes among annotators by making final label decisions and communicated data quality issues to machine learning engineers. I led calibration sessions to standardize labeling practices and authored playbooks to minimize ambiguities. • Audited 5–15% of the general annotation team's output on motion labeling tasks • Implemented a double-blind audit process that reduced dataset error rates by 25% • Hosted "sync" meetings to align the team's interpretation of labeling rules (e.g., bounding box, temporal segment) • Wrote a comprehensive 50-page annotation playbook and standardized guidelines for complex project scenarios.