Projects

Handwritten Digit Recognition Neural Network

A neural network built from scratch in Python to recognize handwritten digits (0-9) using the MNIST dataset. Features a 2-layer feedforward architecture (784→256→10) with sigmoid and softmax activations, trained using mini-batch gradient descent. Includes an interactive web interface where users can draw digits and get real-time predictions. The project demonstrates core machine learning concepts including forward/backward propagation, cross-entropy loss, and model persistence.