Show RN: Tiny scalar-valued autograd engine supporting reverse-mode autodiff in Rust

Why I built this

I wanted to deeply understand how PyTorch and Micrograd calculate backpropagation gradients on a computational DAG, so I wrote a minimal scalar autograd engine in Rust from scratch without external crates.

Key Features

let x = Value::from(2.0);
let y = Value::from(3.0);
let mut z = &x * &y + &x;
z.backward();
println!("dz/dx = {}", x.grad()); // prints 4.0

Feedback on safe Rust memory management for cyclic graphs appreciated!