I've been thinking about developing a tool that would help to transform the graph DB data into a Pytorch dataset and use it with a custom model. Are there some red flags that I should have in mind regarding this tight coupling between ML and DB?