Tensor autograd engine with broadcasting-aware gradients

Extend a scalar reverse-mode engine to NumPy arrays, reduce gradients over broadcast dimensions, and check every operation against torch.autograd.

The exercise happens on Self-directed project (micrograd as the scalar reference). Canonical link: https://github.com/karpathy/micrograd. Link check: link responded (5 Oct 2026). Opened from this app 0 times; opening is not counted as an attempt.

Supports: Deep understanding of the autograd graph and broadcasting for research-engineering coding rounds.