Pocketgrad

* **Robust design models the computational graph using `NamedTuple` dependencies, a standard and effective autograd pattern.** * **Python magic methods enable an intuitive, PyTorch-like API, significantly improving tensor usability and ergonomics.** * **The core `Tensor` implementation is clean, leveraging NumPy for efficient backend array operations and data representation.** * **Testing verifies complex NumPy broadcasting mechanics and accurate gradient flow, confirming mathematical correctness.** * **Abstract the gradient descent loop in `minimize_fn.py` into a dedicated `Optimizer` class for cleaner separation of concerns.**