Quantum Singularity

Up to 110,000 GPU particles arranged into four mathematical topologies, controlled by three fused, real-time input channels: pointer, hand gestures, and live sound.

Point to attract the field, open your palm to control camera distance, or just talk near the mic and watch bass and treble reshape the simulation, all running on-device through MediaPipe Tasks Vision and the Web Audio API. Camera and microphone are strictly opt-in and neither stream ever leaves the browser. Try the live demo, press B for engineering notes or P for a live GPU-time readout.

Gesture recognition fuses a trained neural classifier with a hand-written geometric classifier through a quality-weighted rule, then runs the result through a temporal state machine so a gesture can't flicker on or fail to release from a single bad frame. Every easing and smoothing path (camera distance, gesture hysteresis, audio decay) runs through a frame-rate-independent exponential approach, so behavior is identical at 30Hz and 144Hz. Quality (particle count, pixel ratio, ML inference rate) is tiered at boot from device capability, then adjusted live from measured frame time.

Built and tested across desktop and mobile, with automated Playwright suites validating particle simulation stability, gesture input fusion, audio analysis, and shader compilation.

This was a deep dive into WebGL2 shaders, real-time performance optimization (target: 60fps at 110k particles), and multi-modal input processing in the browser.