This is an educational repository designed to bridge the gap between basic nanoGPT implementations and research-level deep learning. It provides annotated, from-scratch implementations of nearly 100 modern ML techniques including KV caching and speculative decoding for LLMs, vision transformers, attention variants like linear attention, generative models like diffusion and flow matching, and landmark RL papers such as PPO, A3C, and AlphaZero. All code includes detailed comments explaining subtle details typically glossed over in papers and production code.