This repository implements an architecture combining convolutional neural networks (CNN) with linear support vector machines (SVM) for image classification tasks. The project draws inspiration from Y. Tang’s 2013 work on deep learning with SVMs. It replaces the conventional softmax classification layer with a linear SVM while maintaining the CNN feature extraction backbone. The model was evaluated on MNIST and Fashion-MNIST datasets, achieving approximately 99.04% test accuracy on MNIST compared to 99.23% for the CNN-Softmax baseline. Implementation uses TensorFlow.