Most of what I've built follows the same shape: take something messy — a pen stroke, a leaf photo, wrist-sensor noise, a single biomarker curve — and pull the structure out of it.
Sometimes that's a neural net with the governing differential equations baked into its loss. Sometimes it's a greedy nearest-point matcher that turns scribbles into LaTeX. Sometimes it's a 500-gram robot deciding when to charge.
I'm a first-year CS student at Georgia Tech on the Intelligence and Systems & Architecture threads, mostly found in a research lab, a robotics bay, or a restaurant I haven't tried yet.