Your team wants repeatable library code for transformations across jobs and clusters. What packaging and deployment approach is most maintainable on Databricks?

Build a Python wheel in your CI pipeline, publish it to an internal artifact repository or DBFS, and install it on clusters via %pip install or as a job cluster library. This approach gives you versioned, testable code that can be reused across notebooks and jobs. The wheel can be pinned to a specific version, ensuring reproducibility. Using CI/CD to build and publish the wheel automates the process, while Databricks’ library installation mechanisms keep the runtime environment consistent. This is more maintainable than zipping the repo or copying code into notebooks, as it separates code from execution and supports dependency management.