Stage 1 of 10 · 8 lessons · ~4 weeks

From notebook to package

See the whole lifecycle once, then turn prototype LLM code into a Python project with one module for model calls, locked dependencies, config and secrets outside the code, tests and a container image.

Start: The lifecycle and the tool map

Lessons· one lesson for each topic, in this order

Do this

Move notebook code into a src package with one module per job.
Declare dependencies in pyproject.toml and commit uv.lock.
Move paths and settings to a config file and secrets to environment variables.
Add pytest tests for feature or prompt-building code plus one smoke test.
Make training or the LLM feature run with a single command.
Build a Docker image and run the command inside it.

Start with· follow one of these

CoursePaid
Grokking Python Fundamentals
Design Gurus · Beginner · $50
If your Python stops at notebooks: modules, functions, classes and error handling done properly
CourseFree
Made With ML
Goku Mohandas · Advanced
Free: the scripting, CLI, testing and documentation lessons turn notebook code into a real project

Practise

DevOps with Docker
Free exercises that make you build, run and debug images until Dockerfiles feel routine

Keep handy

uv Documentation
Projects, lockfiles, uv run and the Docker integration guide
pytest Documentation
Fixtures, parametrised tests and markers for slow tests

Also worth your time

Packaging Python Projects (opens in a new tab)Read
The official tutorial on project layout and pyproject.toml, the format every tool now reads
Test-Driven Development with Python (Obey the Testing Goat)Go deeper
A free book that makes testing a habit rather than an afterthought

Before moving on: A Git repository where your notebook code lives in a src package with pyproject.toml and uv.lock, config in a file, secrets in environment variables, passing pytest tests, and a Docker image that trains the model or runs the LLM feature with one command.

Tauseef's advice

This stage feels like housekeeping, and it is the one most ML learners skip. Do it properly once and every later stage gets easier, because pipelines, CI and serving all assume code that installs cleanly and runs with one command. Keep your notebooks, but make them import from the package.