Blog
Notes on WebRTC, running machine learning models in the browser, iOS apps and real-time displays. Written while building the projects they come from, not researched afterwards.
Daily weigh-ins swing by pounds while real change creeps along at a tenth of one. The smoothing engine, what a 0.10 constant actually costs you in lag, and the caching layer that fixed a stuttering scrub.
Read itA browser cannot see its own network, so discovery is keyed on public IP. Then two water marks stop a fast sender drowning a slow receiver, and a base64 relay covers the networks that block WebRTC outright.
Read itGetting WebAssembly threads on a static host, picking a model tier from the device, resizing before inference so phones survive it, and an erase and restore brush built on two canvases.
Read itTravelling 62 positions around the spool takes two seconds because only the landing is animated. The four layers of a cell, one frame loop for the whole board, and why every click is detuned.
Read it
