Joel Grus live codes a neural network library in Python.
Grus is a data scientist and presenter of the Adverserail learning podcast. In this YouTube video he builds a neural network library from scratch. It's really useful to see this - both for coding tips and understanding how neural networks work. He builds a basic library, including all of the key elements of a neural network, and then uses it to solve some simple problems.
It's an impressive feat and demonstrates a deep level of understanding: if you can live code something complex and teach it well you must really know what is going on.
It's also good to see somebody actually coding (along with all the mistakes everybody makes) and how he uses the tools (VS code, mypy, extensive use of type hints). You don't get this in standard static tutorials or blog posts. You will learn more in one hour watching this video than many long online courses that teach a sanitised version of the world.