cvat-ai/cvat ? reverse-engineered prompt

Build me a self hosted web app for creating computer vision training datasets. I want users to upload images, videos, and 3D point cloud data, create projects and labeling tasks, define labels, then annotate with boxes, polygons, masks, keypoints, cuboids, and tags.

Make it work for teams, with accounts, organizations, roles, task assignment, comments, issue tracking, review steps, and progress tracking. Add ways to import and export common dataset formats like COCO, YOLO, Pascal VOC, KITTI, and CVAT XML. It should also support connecting cloud storage and have an API or SDK style interface so workflows can be automated later.

Include AI assisted labeling hooks so we can connect our own detection, segmentation, pose, or tracking models to speed up annotation, but keep the first version focused on a solid manual labeling experience. Set it up so I can run everything locally with Docker, create an admin user, open it in Chrome, and start labeling at localhost. Look up current docs online if you need to.