The demo moment I did not plan: I asked my app “how many doors and windows are there?” and the AI answered the counts, then added, unprompted: “note D3 is only 300mm wide, likely a mis-detected door. want me to check it?” It was right. My extraction pipeline had turned a piece of geometry into a 30-centimeter door, and no human had noticed.

That moment sits on top of a long pipeline. It starts with DWG, AutoCAD’s proprietary binary format and one of the most hostile file formats in existence. From it I extract thousands of anonymous line segments, then reconstruct actual architecture: walls from segment clusters, and rooms as faces of a planar graph. The building model is stored as rows, not a blob, so every wall, door, and room is addressable. The AI gets tools to inspect the drawing, move walls, modify rooms, and query quantities, with edits going through the same undo stack as human edits.