Hey, I’m Michael Batrakov, and I’m really excited to share one of my most ambitious projects so far: a Neural Network Visualizer built with Next.js. This project brings together three of my biggest passions — Machine Learning, Data Visualization, and Web Development — into one interactive experience.
What sets this project apart is that I built the forward propagation entirely from scratch in TypeScript, showcasing my expertise in data structures and low-level algorithm design — no external libraries involved. I created this platform as both a learning tool for myself and a resource for others to grasp the inner workings of deep neural networks.
For model v4, a handwritten digit recognizer, I even collected all the data myself — no downloads or external datasets were used. My skills in algorithms are further highlighted through data augmentation techniques, where I shifted digit positions within a 10x10 matrix to expand the dataset and improve training. For a closer look at the process, check out the detailed steps in the Training Link.
Ultimately, this project is more than just a visualization tool — it’s a way to demystify deep neural networks. I hope it inspires others to dive deeper into machine learning and see the beauty in these algorithms!
P.S. If you have any questions or concerns send me an email at mishou@g.ucla.edu
P.P.S. Feel free to connect with me on LinkedIn
Try Model V1