python

Patterns for structuring Python projects managed by uv, with recipes for ML tracking, distributed compute, GPU kernels, and service architecture. Conventions apply to standalone packages and multi-package workspaces.

When to load recipes

You are working on…	Load recipe
uv project setup, deps, or monorepo workspace	`references/recipes/uv-workspace.md`
Publishing a package to PyPI or private registry	`references/recipes/uv-publishing.md`
MLflow experiment tracking or model registry	`references/recipes/mlflow.md`
Ray distributed training or remote tasks	`references/recipes/ray-training.md`
GPU kernels with Nvidia Warp	`references/recipes/warp-hpc.md`
Pydantic models or dataclasses	`references/recipes/pydantic.md`
Click CLI commands	`references/recipes/click-cli.md`
Module architecture, error handling, typing	`references/recipes/service-patterns.md`

Load uv-workspace for any new Python project — every other recipe assumes a uv-managed source tree.