Ramakm/ai-hands-on

Covers mathematics foundations through transformer architectures and production systems (RAG, OCR) using PyTorch, with notebooks implementing neural networks from scratch including attention mechanisms, normalization techniques, and optimizers like Adam and Muon. Integrates with embedding models and vector stores for RAG pipelines, and includes traditional ML implementations (scikit-learn, XGBoost, LightGBM) alongside modern deep learning frameworks.

1,068 stars.