Andrej Karpathy released yesterday great explanation on neural networks that needs to be shared as much as possible.
This kind of low-level (down to the nuts and bolts) approach is also taken by Paolo Perrotta in Programming Machine Learning: From Coding to Deep Learning. Due to the focus on the fundamentals, these sources give (more) confidence prior to entering into popular ML frameworks.
The code is public and can be found here (notebooks) and here (micrograd).