Andrej Karpathy · on YouTube
Andrej Karpathy builds neural networks from scratch in code, starting with backpropagation and micrograd and ending with a GPT transformer and BPE tokenizer. For programmers comfortable with Python and intro calculus who want to truly understand how language models work under the hood.
Developers and technically curious learners. This is an intermediate course in Generative AI & LLMs, running about 8 lectures (~15+ hours), ongoing.
Free. No payment required at any point.
A large catalogue of runnable notebooks, each isolating one retrieval-augmented generation technique, from basic chunking and query transformations to reranking, hierarchical and graph-based retrieval, and evaluation. It suits developers who have built a basic RAG app and want to systematically try the techniques that improve retrieval quality.
Covers the full LLM lifecycle from transformer architecture and pre-training through instruction fine-tuning, RLHF, and deployment, taught by DeepLearning.AI and AWS engineers. The right course for Python developers with basic ML who want to understand how LLMs are actually built and adapted.
Hugging Face's hands-on course on LLMs and NLP using the Transformers, Datasets, Tokenizers, and Accelerate libraries, from fundamentals through fine-tuning, demos, and reasoning models. Best for Python developers who want to work with open models directly rather than only via APIs.