Write OOP from the beginning or refactor?

Im a Jr Data Scientist that is in the process of having solid SWE foundations. My regular ML workflow is first work on a Jupyter Notebook where its usually messy because all of the experimentation and then after I get what I want I transfer into a clean Python script and usually refactored with OOP so it has more maintainability and more "production-ready", then after I finish the refactor code I run some test s so this refactored cleaner code …

data datascience data scientist experimentation jupyter ml workflow notebook oop process python python script solid swe transfer work workflow