DeconBear/learn-ai

learn-ai is a concise, example-driven tutorial repository that explains core AI/ML concepts with diagrams and runnable code. It contains ~55 articles, 200+ illustrations and 100+ code examples designed to be CPU-friendly so a consumer laptop can run them. Each article focuses on a single concept (20–30 minutes to read), from ML foundations (linear/logistic regression, bias–variance) and classical methods (k-NN, SVM, decision trees, ensemble methods) to deep learning (autograd, optimizers, CNNs, Transformers), NLP/LLMs (BERT/GPT, RLHF, LoRA), vision, multimodal, deployment/quantization, and AI safety. An extensive appendix covers 16 algorithm/data-structure chapters for interview/competition prep. Intended users: students, career changers, software engineers moving into ML, ML engineers wanting theory refresh, heavy LLM users, and interview/competition candidates. Key features: single-concept focus, visual intuition, runnable (NumPy-first) code, multiple recommended learning paths tailored to different goals.