numpy-ml is an educational and prototyping library that reimplements common machine learning algorithms without external ML dependencies. It covers neural network layers (LSTM, CNN, ResNet, attention, VAE, WGAN-GP), Bayesian methods (Gaussian processes, mixture models), sequence models (HMMs, topic modeling via LDA), and reinforcement learning agents that train in OpenAI Gym environments. All implementations prioritize legibility over performance, serving as a starting point for ML experimentation.