I was curious about what was under the hood of the code that drives a lot of the machine learning foundations, so I decided to do a deep dive and build it from scratch. The concept is the same: instead of using the abstracted high level functions, try to build a NN from scratch with only the simplest. And what better way to do that than by doing in a lower level language than what was in the video, in C! I was able to get a max accuracy of 94% on the MNIST dataset, which is pretty good for a first try. Any more training with this model would result in overfitting and tank accuracy.