One of the most painful parts of developing neural networks is dependency management in Python. It seems like Python has reinvented multiple wheels that other languages like JAVA have been merrily rolling along on for multiple years. Ironically, Python packages are actually called wheels. Oh well. In this short post we want to show you our solution to this problem for Deep Learning (DL) projects, where this problem is particularly nasty as you also need to juggle multiple CUDA versions. Note that there are multiple ways to deal with this - this just happens to be the one we like most - maybe you will as well?