{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/ec9fc2aac7fa47efa67414ca17ca7c04\" frameborder=\"0\" width=\"1920\" height=\"1440\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":1440,"width":1920,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":1440,"thumbnail_width":1920,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/ec9fc2aac7fa47efa67414ca17ca7c04-1f7879d16d6e1c02.gif","duration":201.302,"title":"How Analysts Fix AI Detection Errors","description":"This Loom demonstrates Atlass, a triage tool that reviews AI-detected objects in overhead imagery and dispatches confirmed reports. The author shows how a truck depot image can be misclassified by the model at low confidence, while an analyst quickly reviews it in about 5 seconds, logs model failures, and flags images for rescans so mistakes are not silent. They then review another set from Bridge Sector 1 where all 26 images were detected as cars and explain integrated shortcut and one-button dispatch behavior. The PID section covers design tradeoffs, including time-to-dispatch with analyst-confirm precision, and notes the current build timeline of 5 days by one developer using Next.js, YOLOv8, and Supabase, deployed on Vercel."}