Atlas capture
I handle dense atomic action labeling for the Atlas Capture program, a massive, high-volume project training advanced AI models on human behavior and hand-object interactions. My role involves breaking down short, first-person video clips into precise segments, adjusting timestamps, and writing sequential, comma-separated descriptions of physical tasks. To maintain the project's strict quality standards, I had to pass a rigorous 25-question qualifying assessment and must consistently adhere to precise guidelines regarding action-merging rules and detailed object descriptions to clear the platform's consensus checks.