Building a neural network from scratch without any libs, just pure linear algebra and some other cool mathematics. I feel libs make things more complex than it should and everyone working with neural networks should try to build one from scratch using the concepts behind them. Remember that this field is kinda old, and a lot of people worked a lot so we can be here. Its good to understand what have been done, so we can improve it even more in the future, from scratch, raw and in any language. Here we are doing it in rust because we can. And if you take a look at the code, you will see its synthax is cleaner than some crazy pytorch that some C++ addicted programmer wrote when he was high on acid.
Neural networks are simply compositions of neurons. Each neuron computes a weighted sum of inputs plus a bias, followed by an activation function. Everything else in deep learning is built on that same primitive.