Backflip AI attacks a long-standing bottleneck in product development: CAD is powerful but slow and manual, so rebuilding geometry from scans, drawings, or photos can consume hours of engineering time. The company trained what it describes as the first foundation model designed for CAD, one that reasons in construction operations — extrude, revolve, pattern, chamfer, fillet — rather than generating geometry that merely looks right. The output is native parametric geometry with a sensible feature tree, sketch constraints, and external references, exportable as a .STEP file or opened and edited in mainstream CAD packages.
The product ships as three related capabilities. Scan-to-CAD converts 3D scanner output and mesh files (STL, OBJ, GLB, GLTF, PLY) into editable parametric models, notably faster than manual reconstruction. Drawing-to-CAD reads legacy engineering drawings, interprets dimensions and geometry, and rebuilds them as modern 3D models — a way to unlock decades of archived drawings. Photo-to-CAD turns a few phone photos into a parametric model, with one known dimension used to scale everything accurately.
Integration matters for adoption: Backflip currently works as an add-in to Autodesk Fusion plus a standalone web app, with plug-ins for Onshape and other CAD software stated as coming. The company says models are optimized for 3-axis CNC and turned parts today, with sheet metal and injection molding on the roadmap. On the security side it encrypts data in transit and at rest, separates access by organization, does not train on enterprise data, and is working toward SOC 2 Type II. The team is backed by investors including Kevin Scott, Rich Miner, and Ashish Vaswani.
