What is tagcast-ai?

TagCast is a data annotation platform built for teams that need to label training data for AI models. It covers the entire process from raw data to finished labels, including managing datasets, doing the actual annotation work, checking quality, and evaluating how well models perform on the labeled data. The description is written in Chinese, but the codebase is open to anyone who wants to run it locally or on a server. The platform handles five types of data: text, images, audio, video, and 3D point clouds. For each type there is a dedicated annotation workspace where team members can draw bounding boxes for object detection, classify images, answer questions about text passages, label spoken audio, or mark up 3D scans. Labels are organized into reusable libraries and templates so annotators do not have to rebuild the same label sets project after project. Quality control is built in through a separate review step where a second person checks completed annotations and flags inconsistencies. There is also a model evaluation area where you can compare predictions from different models against your labeled data to see which performs best before you commit to training. On the technical side, the backend runs on Python with FastAPI and stores data in PostgreSQL for production or SQLite for local development. Redis is used as an optional cache. The frontend is plain HTML, CSS, and JavaScript with no framework. Everything can be started with Docker Compose, which brings up the backend, database, Redis, and an Nginx reverse proxy together. Access is controlled through three roles: a super admin who manages the whole system, project managers who set up datasets and teams, and annotators who do the labeling work. A separate admin dashboard gives the super admin visibility into users, projects, and system configuration.