How should I structure a multi-language project from scratch?

Learn the basic Git workflow early. For a solo project, add files, commit working milestones, and push them to a remote repository regularly; learning branch, switch, merge, and reset afterward will cover most collaborative situations. Each language generally has its own dependency-management system, so document the required versions and setup steps in a README. If the environments become difficult to reproduce, a container such as Docker can later provide a consistent system, but it’s probably unnecessary for the first proof of concept.

Break the idea into several small pieces rather than treating it as one huge project. Build and test each part independently, then connect them through a simple interface such as command-line arguments, files, or an HTTP API. If the first integration attempt fails, that’s useful information—it will show you what each component needs from the others. Looking at small, well-organized open-source projects or following a tutorial close to your use case can also help you choose a reasonable structure without overengineering.

Start small instead of designing the entire architecture up front. For a proof of concept, you could have a Python entry point that calls the AI API and invokes Lean when needed, perhaps through Python’s subprocess module. Use a Python virtual environment for Python dependencies and Lean’s own tooling, such as Lake, for Lean dependencies. Keep API keys in a local .env file and add that file to .gitignore so secrets never enter version control. Once the basic workflow works, you’ll have a better idea of which parts deserve a cleaner design.