Seventeen year old Greg Tarr from Bandon Grammer School in Cork won the BT Young Scientist & Technology Exhibition (BTYSTE) for creating a more efficient deep fake detector. Check out his video explaining the efficiencies he achieved here.
Chris Wellons on null program describes how he built a race car simulator using polynomial calculations instead of using a more complicated, and this case less efficient machine learning solution, in order to demonstrate that not every problem needs to be solved with machine learning.
Neural Radiance Fields For Unconstrained Photo Collections (NeRF) is a neural network project from researchers at Google that can take in a set of pictures of a place from the internet and automatically generate a representation of the scene that can be explored in 3D. The lighting and environmental effects of the scene can also be manipulated at a realistic level.
Hackster.io has a useful how to guide by Jen Looper describing the steps involved in building a bird identifier using a custom machine learning model running on a Google Coral Dev board with a Raspberry Pi Camera module.