0xhassaan/nn-from-scratch

This repository is a personal learning log. The author is working through a well known online video series by Andrej Karpathy, a former research lead at OpenAI and Tesla, in which Karpathy builds the core machinery of modern neural networks step by step in Python. The point of the project, as the author puts it, is that every line of code has been written by hand and understood, not just copied or run. The README is essentially a checklist of the lessons in that series. One item is marked done: micrograd, a tiny library that implements the backpropagation algorithm, the math trick that lets neural networks learn from examples. The remaining items, all unfinished, cover a series called makemore that builds increasingly capable text generators, starting with a simple bigram model, then a small multilayer network, then techniques like batch normalization and the WaveNet architecture. The final planned item is nanoGPT, a minimal reimplementation of the kind of model that powers ChatGPT. There is no description of features, no install instructions, and no usage notes in the README. The repository is best read as a study journal rather than a finished tool.