Autodiff

A minimal automatic differentiation engine for scalar computations implemented from scratch in Python. Every arithmetic operation builds a directed acyclic graph (DAG) of Value nodes that stores gradients and supports backpropagation through the chain rule. Includes neural network building blocks (Neuron, Layer, MLP), an SGD optimizer, and computational graph visualization with Graphviz. Built to understand what's really happening when you call .backward() in PyTorch. View on GitHub at github.com/Salmankabir12/autodiff.