Ground Truth Curator

A platform for subject-matter experts to create and maintain high-quality ground truth datasets for agent evaluation and model accuracy measurement.

Quick Start

See the documentation for complete guides.

Installation

# Backend
cd backend
uv sync

# Frontend
cd frontend
npm install

Running Locally

# Start backend
make -f Makefile.harness backend

# Start frontend (in another terminal)
make -f Makefile.harness frontend

# Or run both from one terminal
make -f Makefile.harness dev

# Or run both in the background
make -f Makefile.harness dev-up
make -f Makefile.harness dev-down

These targets wrap the existing local dev commands in backend/ and frontend/. Use dev for a foreground session, or dev-up / dev-down when an agent or developer wants background-managed servers with logs in .harness/dev/.

To start the background demo stack with seeded demo data and a stable local identity, run:

VITE_DEMO_MODE=true VITE_DEV_USER_ID=demo-user make dev-up

This enables the demo UI flow, seeds the backend memory repo with demo items, and uses demo-user for assignment-aware API calls. Stop it later with make dev-down.

If you prefer a single shortcut target, make -f Makefile.harness dev-up-demo applies the same memory-backed demo settings and demo-user identity automatically.

Running Tests

# Backend tests
cd backend
uv run pytest tests/unit/ -v
uv run pytest tests/integration/ -v

# Frontend tests
cd frontend
npm test -- --run

Documentation

Full documentation is available at https://andrewvineyard.github.io/GroundTruthCurator/

To build documentation locally:

cd backend
uv run mkdocs serve -f ../mkdocs.yml

Then open http://localhost:8000

Contributing

See CONTRIBUTING.md for development guidelines.

License

[License information to be added]