ML-first scaffold for rapid experimentation, training, and deployment.
- Fast iteration on models with reproducible experiments.
- Clear separation of data, training, serving, and ops.
- Infrastructure-ready structure for productionization.
docs/ Architecture and planning docs
data/ Data contracts, schemas, samples
models/ Model artifacts and checkpoints (gitignored)
notebooks/ Exploratory notebooks
scripts/ Utility scripts for data and training
src/ Application and ML code
infra/ Deployment, CI/CD, and infra configs
tests/ Automated tests
- Copy
.env.exampleto.env. - Create and activate virtualenv.
- Install package with dev dependencies:
pip install -e .[dev,ml]
- Train baseline model:
python -m src.train
- Run API:
uvicorn src.serve.app:app --reload
- Execute tests:
pytest
See docs/next-steps.md for a practical roadmap (2 weeks, 6 weeks, 3 months).
- Backlog wykonawczy:
TASKS.md - Roadmapa etapów:
docs/next-steps.md
- Training automatycznie loguje parametry, metryki i artefakt modelu przez MLflow (jeśli extra
mljest zainstalowany). - Gdy MLflow nie jest dostępny, trening działa dalej z fallbackiem (Noop tracker).
- Szybki start agenta:
./scripts/setup_agent.sh - Instrukcja auto-update klienta:
docs/agent-setup.md
## Quickstart (suggested)
1. Copy `.env.example` to `.env`.
2. Create a virtualenv and install dependencies.
3. Run a baseline training job: `python -m src.train`.
4. Serve the model: `uvicorn src.serve.app:app --reload`.
5. Start with `docs/architecture.md` and `docs/plan.md`.