I made a proof of concept application that made it easy to combine different opencv image filters to make a filter pipeline. The neat part was the genetic algorithm on top that "tuned" the filters for best results for some task (finding circles in my test case).
Currently, I'm working on integrating multi-sensor arrays into brick and mortar retail for analytics and interaction purposes. I'm also hiring, so if you have experience with ML and / or multi-sensor arrays / HCI, ping me. Email in profile.
Coming up with an algorithm to implement the "as if" rule for a C compiler, to avoid needless widening. (Not my current job, I hasten to add.)