Neural Network Visualizer

The Problem

Neural network architecture is notoriously difficult to communicate. Existing tools were either too academic for practical use or too simplified to be informative. There was no lightweight, browser-based option that showed weight propagation, layer structure, and activation functions interactively without requiring installation.

The Approach

Built a browser-based interactive visualizer that renders configurable neural network architectures in real time. D3.js handles the SVG layer graph. The Python backend generates the initial weight matrices, exported to JSON for the static frontend to consume.

Static Vercel deployment — no server, no runtime cost, loads instantly.

The Outcome

Used as a reference and teaching tool. Demonstrates machine learning depth beyond what a typical web developer portfolio communicates — the ability to work at the intersection of data science and frontend engineering.