Neural network library built from scratch in Java, no external math were used, all of math calculations were done manually including there vectorization steps The Neural Network class is built from scratch containing activation layers and Linear Layers, quite simple

Then there is a frontend to just showcase the app, it's a mobile app built in Java where the user can connect to the backend server, then draw a number and send it to the backend the backend then processes the image through the Neural Network model it got and returns a response containing what is the number it predicted

Note that training was not done in Java but rather in python due to many reasons (mostly due to GPU integration) so after training was done in python using Pytorch the weights were imported to the Java end