From Floor Plan Photo to Interactive 3D Model: A Systems Perspective on Production AI for the Built Environment

Converting a 2D floor plan image into a navigable 3D model is a long-standing problem incomputer vision and computer graphics, with applications in real estate, interior design,accessibility, and consumer 3D printing. Most published work focuses on academic benchmarks(CubiCasa5K, R2V, RPLAN) or assumes inputs that have been pre-cleaned. In contrast, real usersupload phone-camera photos of paper printouts, multi-page architectural PDFs, screenshots ofreal-estate listings, and noisy scanned blueprints. We describe Ritn3D, a deployed productionsystem that takes such inputs from iOS, Android, and web clients, runs an asynchronousdetection-and-rendering pipeline backed by a headless Blender renderer, and returns a shareableinteractive 3D model with optional GLB/STL export. We report public, founder-stated accuracyfigures across four input categories: 90–95% on CAD-exported PDFs, ~80% on clean scannedblueprints, 75–85% on phone-camera photos, and below 30% on photo-realistic real-estate listingrenders. We discuss the systems-level decisions (asynchronous render queues, deterministic ratherthan generative rendering, human-in-the-loop correction UI, regional pricing) that make theproduct viable at a single-founder scale, and document five honest production lessons. To closethe academic–production gap, we open-source ritn3d-floorplan-eval , a PyPI-installableevaluation harness for floor-plan detection, and ritn3d-stl-tools , a toolkit for cleaningBlender-exported STLs for 3D printing. The paper is intended as a transparent post-mortem of ashipping product, not a novel-architecture contribution.