MMSmartShape3D

Turns flat 2D steel drawings CAD files into fabrication-ready 3D models and machine files in seconds. $14,000 revenue in 90 days through an industry dealer. Built with Gemini, run on Google Cloud.

Comment

What it does

Steel fabrication shops run laser cutters that need 3D models and machining data — but factory drawings arrive flat, as 2D CAD files (DXF). So today a person rebuilds every part by hand, and people make mistakes. I found this problem by talking to factory workers in Korea, and I automated it: drag a DXF in, and in about four seconds MMSmartShape3D reads the drawing, stands every part up in 3D, and writes the files the cutting machines actually read. Every hole, cut, and marking comes through exactly as drawn.

That is why this entry is filed under Professional Services: the product replaces the manual drawing-to-fabrication engineering work that shops currently pay for as skilled labor.

Try it — nothing is a mockup

Business Viability

  • $14,000 in revenue in the last 90 days, from real customers, sold through an industry dealer.
  • The model: the web demo proves the conversion works, free. The production work happens in the paid desktop app, whose licenses come from a license server I built on AWS — machine-bound seats, atomic activation, RSA-signed responses, and a 7-day offline grace period.
  • The dealer sells the software bundled with training and support; that channel is where the revenue comes from.

AI-Native Operations

  • Gemini in the product's production path: Gemini handles the ambiguous profiles. It reads the text on the drawing — and only the text, never the geometry. If the text can't decide, the geometry does. Every model call is recorded in the Gemini usage log.
  • AI-built: the product was vibe-coded 100% with Antigravity and Gemini Flash — which is how one person shipped it this fast.
  • Agents run the operations: a support-chat agent answers first questions and escalates the serious ones to me; a lead-briefing agent reads what a prospect converted and briefs me before we meet. Every decision and every model call is logged on one ops dashboard.
  • The honest split: this company doesn't run 100% on AI, and I should say that. Face-to-face sales, training, and consulting stay human — by design. The rest is AI.

Accuracy — measured, not claimed

  • 339 parts verified against their source drawings; largest deviation 9 microns; zero failures.
  • 32 regression tests run on every build.

How I built it

A Python desktop application (the paid product) plus a web conversion service deployed on Google Cloud Run, with the Gemini API in the deployed conversion path. The license server runs on AWS. The support-chat agent, lead-briefing agent, decision log, and Gemini usage records are all in the repository and visible on the ops dashboard.

Built With

Share this project:

Updates