I built SVGotchi, an SVG-native local AI companion where the opened SVG document is the app surface. There is no Python backend, no hosted inference API, and no normal HTML app shell. Transformers.js loads a local ONNX emotion model in the browser, then TypeScript maps sanitized labels and scores into deterministic SVG pose transitions. I’m looking for feedback on the architecture, README positioning, and whether the SVG-native runtime boundary makes sense as a reusable pattern for small local...