Changwang Yi

Autoregressive sequence models can enforce strict multi-view CAD geometric alignment without raster vision encoders, simply by restructuring SVG syntax to establish local coordinate spaces before semantic boundaries.