I recommend you give Panda3D a try, particularly the C++ API. Here's an example of my musical work visualized in Panda3D: https://www.youtube.com/watch?v=bHW7B2TKk6Q
> Do you have any tips on achieving demoscene type graphics and music?
I think you should reconsider your line of thought. demoscene type graphics and music are usually coded with assembly for speed optimization and there are hard limits on the size of the binary, usually.
Is that really what you are pursuing?