This is exactly what we do, but with Spark instead. We develop the functions locally in a package and call necessary functions for the job notebooks, and the job notebooks are very minimalistic because of this