Founding AI Engineer, Scan-to-BIM

We are empowering people to go from concepts to construction within hours, enhancing collaboration in the world of atoms & pushing the boundaries of what’s possible, for humanity over the next few decades. For phase 1, we're training foundational AI models that create building code-compliant BIM models, cost analysis & documentation from 2D building blueprints. We are a small team of 20+ and looking for people passionate about 3D generation/reconstruction. We’re targeting $100B+ opportunity in construction’s design, estimation, and planning software—ripe for disruption by AI-first platforms. If you're interested in joining an early stage AI startup building foundational models, geometric and physics engines from scratch to transform how the world designs and builds infrastructure, with real autonomy and the resources to solve hard problems, apply or reach out to us today. About the role Augrade is building an AI-native Digital construction platform to automate how the physical world is modeled, estimated, and constructed. We are moving beyond manual & tedious workflows towards fully automated building generation from raw 2D or spatial data using AI. As a Founding AI Engineer for Scan-to-BIM on our core AI/3D team, you will own the problem end-to-end: from raw 3D data to semantically rich building models. This is not a BIM modeling role. It is a systems and algorithms role focused on building the pipelines that make BIM modeling obsolete. You have built real systems on 3D spatial data — point clouds, meshes, or volumetric representations — and you understand the gap between academic reconstruction benchmarks and messy, production-scale building scans. You have debugged geometry pipelines, wrestled with coordinate systems and registration drift, and shipped something that ran on data you did not control. What you'll own End-to-end pipeline ownership. You design and ship the full scan-to-BIM stack — ingestion, registration, semantic segmentation, geometric reasoning, and parametric model generation. You set the accuracy bar and you own it when it breaks. 3D scene understanding at building scale. Build and improve algorithms for classifying, segmenting, and reconstructing building elements — walls, floors, ceilings, MEP — from real-world, noisy, incomplete scan data. You're not running benchmarks. You're solving problems that benchmarks don't cover yet. BIM generation, not BIM consumption. Translate model outputs into IFC-compliant, Revit-compatible structures that AEC teams can actually build from. Own the bridge layer between AI predictions and production BIM — understand both the geometry and the downstream contractual requirements. Define what "correct" means. Build the evaluation frameworks, accuracy benchmarks, and validation pipelines that measure not just model performance, but constructibility and real-world usability. If there's no metric for it yet, you create it. About you Hands-on experience with 3D computer vision, LiDAR/point cloud processing, or large-scale geometric reconstruction — in production, not just research. Strong Python and/or C++; you can architect and ship a pipeline end-to-end, not just contribute components. Working knowledge of Open3D, PCL, or equivalent 3D processing frameworks. Ability to move fast in ambiguous problem spaces — first principles thinking, not playbook execution. Nice to have Experience with semantic or instance segmentation of 3D data (e.g., walls, floors, ceilings) Exposure to graph neural networks or learning on 3D/graph-structured data Background in digital twins, reality capture, or building reconstruction systems Experience with procedural or parametric model generation Prior work bridging AI outputs with CAD/BIM tools Benefits Generous PTO / sick leave ESOPs (commensurate with impact) Visa sponsorship (for select candidates) More as the team grows