Neural network to classify handwritten digits.
This demo implements a vanilla feed-forward neural network from scratch in order to classify handwritten digits from the MNIST dataset. Users can experiment with different network parameters, including the number of hidden layers, neurons per layer, learning rate, and batch size, to observe their impact on classification accuracy. Training and inference are executed entirely in the browser using Web Workers, ensuring that intensive computations do not block or degrade the responsiveness of the user interface.
Tool to edit, trim, or crop MCAP files.
An intuitive, responsive interface designed for processing large robotics .mcap files. By reading and writing directly to disk, the tool can process files of any size without memory constraints, limited only by available storage. File operations run at near-native speeds via WebAssembly, while dedicated Web Workers execute heavy background processing to keep the UI smooth and responsive. To prevent data loss, cropped exports are streamed safely to OPFS first, then files can be saved to the user's desired location on their file system.
Real-time audio frequency analysis.
The spectogram output is displayed as a rolling heatmap of frequency signals, with decibels (dB) represented according to a colour pallete, time on the vertical axis, and frequency on the horizontal axis.
The frequency analyser output is displayed as a graph, with decibels (dB) on the vertical axis and frequency on the horizontal axis. Peak values remain visible until manually reset.
Path finding algorithm visualizer.
This demo implements a visualizer to compare different pathfinding algorithms with mutliple heuristic options. Users can interactively place start and end nodes, draw obstacles, and observe how each algorithm explores the grid. The visualization highlights the nodes visited, the final shortest path, and the performance differences between algorithms in terms of speed and efficiency. This tool is useful for understanding the mechanics and trade-offs of each algorithm in real time.
Interactive puzzle games.
A React-based implementation of NYT puzzle games designed to feel smooth and responsive. It focuses on a clean, easy-to-use interface and a consistent experience across devices. The games are fully polished with fluid animations that provide intuitive user feedback as they are played.
Want to discuss a collaboration or proposal? Fill out the form to get in touch.