- Built a scalar-valued autograd engine in Rust from scratch, implementing reverse-mode automatic differentiation with a dynamic computation graph. - Leveraged Rc> for shared mutable state, enabling gradient propagation through closures stored on each node while satisfying Rust's borrow checker. - Implemented operator overloading for arithmetic ops and tanh activation with correct gradient backpropagation via topological sort over the computation DAG. - Constructed a full MLP framework supporting configurable layer depths and neuron counts; trained a 2-4-4-1 network to binary classification in 200 epochs using MSE loss and gradient descent.