This Buddhabrot WebGPU Explorer aims to be a high-performance fractal renderer built with Three.js and the Three Shading Language (TSL). Unlike traditional escape-time fractals, this application implements a probabilistic density-mapping algorithm that tracks the entire orbital trajectory of points outside the Mandelbrot set, revealing the iconic "ghost" figure hidden within the chaos. To ensure true non-determinism, the simulation is seeded with verifiable quantum randomness from the CURBy beacon, derived from Bell-certified quantum entanglement. Key technical features include WebGPU compute shaders for real-time path accumulation, 4D Juddhabrot rotation through higher-dimensional space, and a multi-spectral Nebulabrot mode that maps different iteration depths to discrete RGB channels.

This project is a technical exploration of the future of web graphics, featuring six procedural planets built using WebGPU and Three.js Shading Language (TSL). By shifting away from traditional textures, I leveraged TSL’s node-based architecture to experiment with real-time vertex displacement, custom atmospheric lighting, and complex noise functions. Each planet serves as a unique case study in high-performance shader logic, demonstrating how next-generation GPU APIs can be used to create immersive, mathematically-driven environments directly in the browser.

This is a lightweight neural network classifier built with Vite and powered by Brain.js, designed to predict whether mushrooms are edible or poisonous based on their physical characteristics. It uses the UCI Mushroom Dataset, which contains 8,000+ samples of gilled mushrooms from the Agaricus and Lepiota families, each described by 22 categorical features such as cap shape, odor, gill color, and spore print color. The project converts these categorical features into one-hot encoded input vectors suitable for training in Brain.js. A feedforward neural network is configured with a single hidden layer and trained entirely in the browser, leveraging Brain.js’s JavaScript-based GPU acceleration when available. Once trained, the model can classify new mushroom samples with reasonable accuracy and immediate feedback. Everything runs client-side with zero dependencies beyond Brain.js and Vite, making it fast, interactive, and easy to experiment with directly in the browser.

This project presents a straightforward implementation of Dijkstra's algorithm for calculating the shortest path on a mesh surface using Three.js. Leveraging Vite for rapid development, Three.js for 3D graphics, and React-Three-Fiber for seamless integration with React, it creates an interactive visualization where users can explore and understand the algorithm's functionality in real-time.

Earth Data Visualization is a proof-of-concept project developed during a hackathon. It offers a simple tool for visualizing data usage on a globe. It uses datasets with geographic coordinates, and the tool plots the data points on the Earth globe accordingly. It provides a straightforward means of exploring data geographically within a limited scope. It was created using three.js for the 3D graphics and Next.js as a framework.

This project is an experimental blend of Three.js and GLSL shaders, designed to generate dynamic particle effects upon hovering over images. It originated as a variation of @brunoimbrizi's Interactive Particles repository and is featured in his Domestika course, Creative Coding 2.0 in JS: Animation, Sound, & Color. The demo showcases a series of Buddhist mandalas, offering users the opportunity to interact with them by simply hovering over each image.

This happened as part of an accident while playing with shaders some years ago, trying to create a geometry displacement shader that would become transparent on hover. The result was this sort of black hole into hell, which I thought looked quite interesting.