fast.ai

fast.ai takes a deliberately top-down approach — you train a working deep learning model in the first lesson, then work backward into the theory, rather than building up from linear algebra and gradient descent first. The accompanying fastai library wraps PyTorch with sensible defaults that get a reasonable model working with far less boilerplate than raw PyTorch requires.

Aimed specifically at people who want to build things quickly and learn the “why” as they go, rather than a traditional bottom-up ML course — a genuinely different pedagogical bet, not just another PyTorch tutorial.