How do you set up dependencies and environments on a new Python project?

Declare everything in pyproject.toml and manage the environment with uv, which creates the virtual environment, resolves dependencies, and writes a lock file quickly. Keep direct dependencies loose in the project table, commit the lock file for reproducible installs, and separate development tools into dependency groups so production images do not ship pytest and ruff.

Everything lives in pyproject.toml: the dependencies, the tool configuration for ruff and mypy and pytest, the build metadata. I use uv to drive it, because it creates the venv, resolves, and installs fast enough that a cold CI environment is seconds rather than minutes, and it manages the interpreter version too so nobody is fighting over which Python is on the path. Direct dependencies get reasonably loose constraints, then uv lock produces the exact resolved set and that lock file is committed. CI and Docker install from the frozen lock, so what runs in production is bit for bit what I tested. Development tools go in a separate group, which keeps the runtime image smaller and means a compromised linter is not sitting in the production container. In Docker I copy the lock and manifest first and install before copying source, so layer caching actually works. Migrating an older project off pip and requirements files took maybe an afternoon and cut our CI setup step by several minutes.