Do you mean something like Netflix Vizceral[1]? The problem with using data is that there's usually too much of it, and at that point you do end up with a graph that only something like graphviz can layout.

I know what you mean though; but it has to be IMO just enough data, that can then be tweaked by a human to help with the key purpose of such diagrams - human understanding.

1: https://github.com/Netflix/vizceral