I cannot overstate how AMAZING this is! We sorely need more tools to ease mathematical intuition. As far as we've come with scientific computing we still can't feel and touch our math easily.
I looked into implementing visual programming for my work a few years back but figured it'd be an unrealistic amount of work given the state of visual programming libraries in Python (I'd looked at PyFlow back then but it wasn't quite mature enough). I'm definitely interested in contributing. I have experience in both numpy and pygame and I do scientific computing for a living.
Since you accept feature requests, one tool I've been looking for is something to implement and visualize partial differential equation systems easily. It's basically impossible to program cleanly in Python. There are systems built from dozens of equations that each are expressed by something like 200 characters per line of code that look something like:
x1 ** 3.4 + (PA / P**(3x2 + z0)) * V**t - y + (z + (7 - De/4) * 5) * DVe - ...
It's basically unreadable, un-debuggable, it's a pain to implement, a pain to iterate over and good luck to whoever comes after to try and understand what it does. And you might try to use the Object-Oriented paradigm to make it more modular and understandable with verbose functions and variables but in the end it still doesn't work. Matlab's got it working somewhat with simulink & Cie but who wants to use matlab these days. Your work already goes a long way with regards to this problem but what would bring it over the top for me would be the ability to render a block system into a latex equation so I can compare my implementation directly with the paper it came from. The second would be to have aliases for functions, some sort of metadata, so that functions can have verbose names and their associated mathematical symbol (velocity -> v).