Hand-Written Digit Classifier

My first real machine learning project: training a neural network to recognize handwritten digits from the MNIST dataset. It started as a course assignment but turned into a deeper dive into how neural networks actually learn. I built the architecture from scratch, trained it on 60,000 images, and got it to over 95% accuracy on test data.

Technical Details:

• Multi-layer neural network built from scratch
• Trained on 60,000 images from the MNIST dataset
• Achieved over 95% accuracy on test data
• Backpropagation and gradient descent optimization