DEvol (DeepEvolution) uses genetic algorithms to automatically search for optimal neural network architectures. Each candidate network is encoded as a fixed-width genome describing layer types, activation functions, dropout rates, batch normalization, and pooling settings. The tool evolves populations of architectures over generations, applying crossover and mutation to improve model performance on classification tasks. It targets Keras models and was demonstrated achieving 99.4% accuracy on MNIST without manual architecture design.