semant_text_cl_app

Human-in-the-loop text classification app for creating benchmark labels.

Stack

• Backend: FastAPI, SQLAlchemy async, fastapi-users, Pydantic v2
• Frontend: Quasar (Vue 3 + TypeScript)
• DB: SQLite by default (DATABASE_URL can point to PostgreSQL)

Backend (dev)

cd backend
pip install -r requirements.txt
PORT=8005 ALLOWED_ORIGIN=http://localhost:9005 python run.py

Backend defaults to http://localhost:8002 if PORT is not set.

Frontend (dev)

cd frontend
npm install
API_URL=http://localhost:8005 quasar dev -p 9005 -h 0.0.0.0

API_URL is injected by Quasar build config (build.env) and used by axios boot file.

Frontend build with custom backend URL

cd frontend
API_URL=http://api.example.com:8005 npm run build

The compiled app will call that API URL at runtime.

Main API

• GET /api/tasks
• GET /api/tasks/{task_id}
• POST /api/texts/next
• POST /api/annotations
• GET /api/stats/me
• GET /api/stats/leaderboard/{task_id}
• GET /api/admin/tasks
• POST /api/admin/tasks
• PUT /api/admin/tasks/{task_id}
• PATCH /api/admin/tasks/{task_id}
• POST /api/admin/texts
• POST /api/admin/tasks/import-prompts

Notes

• Task definitions are stored in DB and support single-choice or multi-choice labels. Each class may carry an optional description shown as a tooltip to annotators.
• Text uploads are JSONL; full row JSON is stored for downstream benchmarking/export. Individual texts can be suspended by admins to exclude them from the annotation queue.
• Users may set a display_name at registration; this name appears in the navbar and on leaderboards.

Docker compose deployment

cd deploy
docker compose up --build

• Frontend: http://localhost:9000
• Backend: http://localhost:8002

Use deploy/frontend.Dockerfile build arg to set API URL:

docker build -f deploy/frontend.Dockerfile --build-arg API_URL=http://localhost:8002 -t text-classifier-frontend .