At one point, I did a similar thing for a computer vision system I was working on.

That was quite naturally a filter graph: every component was time-dependent, caching and logging was important (rewind to find this blob 5 minutes ago), and parallelism was crucial because the system had to deal with synchronizing multiple IP cameras.

Amazingly, it was possible to do this in an entirely safe way in C++, with templates.

I wish I could put the code up on github, but unfortunately the company got taken apart and sold, so the code is in limbo.

This is great work. Stuff I've been thinking about for a number of years but never actually doing. I wish there were something like this for Python. I know it's not pure, but it's my drug of choice. Anyhow, kudos.

Coauthor of Graph here, I'm happy to answer any questions.