data annotation
Led video annotation for a home cleaning/physics AI training dataset. Annotated 150+ hours (85,000+ frames) of 4K household cleaning footage across 12 real home environments. Object detection & segmentation of cleaning tools/surfaces, action recognition with temporal labeling, motion trajectory tracking, physics-based annotations (velocity, force, fluid dynamics), and human pose/interaction labeling. Quality Measures: Achieved 96%+ inter-annotator agreement (IoU > 0.85), multiple QA reviews, and strict guideline adherence. Delivered high-quality labeled data for computer vision and robotics model training.