MLOps boilerplate bringing software rigor to Python ML pipelines

What it does

This repository is an opinionated Python template for MLOps workflows. It glues together a modern stack—uv, ruff, mypy, pytest, mlflow, pandera, and docker—into a single project structure. You define jobs like training or inference in YAML files, and the code dispatches them using Pydantic discriminated unions. It is essentially high-quality glue code: it enforces software-engineering standards without pretending to be a new framework.

The interesting bit

The clever angle is treating MLOps as a config-driven software package rather than a tangle of scripts. Pydantic validates both your data schemas via Pandera and your runtime job selection via discriminated unions, so the same codebase handles training, tuning, promotion, and inference without branching into spaghetti. It is a rare template that remembers ML code still needs type checking and unit tests.

Key highlights

• Heavy use of modern Python tooling: uv for environment management, ruff for linting/formatting, mypy for typing, and pytest with parallel execution.
• Configuration-driven execution via YAML and OmegaConf, with Pydantic discriminated unions automatically routing to the correct job task.
• Integrated MLflow for experiment tracking, model registry, dataset lineage, and system metrics collection.
• Data validation through Pandera schemas on Parquet-backed Pandas containers.
• CI/CD via GitHub Actions and packaging as both a wheel and Docker image.

Caveats

• You must adapt the code to your specific compute platform or registry; it is a starter, not a turnkey platform.
• It pins you to a very modern stack—Python 3.13 and uv—which may exclude teams on older toolchains.

Verdict

Worth a look if you are starting a new MLOps codebase and want software-engineering guardrails from day one. Skip it if you already have a settled stack or need a framework rather than a structured template.