Instant Custom Jewelry: A Vision + LLM pipeline that sizes rings, generates CAD, and prices in under 90 seconds

I’ll demo a working pipeline that turns a quick hand scan into a priced, manufacturable ring design—live, no slides.

Flow (end-to-end): (1) WebRTC camera feed captures the customer’s hand; (2) a small CV model + OpenCV infers ring size from finger landmarks and scale cues; (3) a JSON spec is composed via an LLM prompt (style, metal, stone size, budget); (4) parametric CAD is generated (OpenSCAD) and exported to STL; (5) a geometry pass computes volume/weight for price; (6) a Shopify draft order is created with thumbnails and options.

What I’ll show: latency breakdowns, prompt templates, failure modes (skin tone/lighting, hand pose), and tradeoffs between on-device vs. API inference. I’ll open the repo and walk through the exact modules so builders can replicate this for any mass-customization retail flow.