This is like giving your CAD/design software a smart AI co-pilot that can propose and refine 3D shapes by itself, while a human designer stays in the loop to guide and approve each step.
Traditional geometric and parametric modelling in construction and industrial design is slow, expert‑heavy, and highly iterative. This work proposes an AI‑agent workflow driven by large models to automate much of the routine exploration and editing of geometric designs, while still allowing human experts to supervise and steer the process.
If productized, the moat would come from proprietary datasets of geometric models and design histories, plus tight integration into existing CAD/BIM workflows that make the AI agent hard to replace once embedded.
Hybrid
Vector Search
High (Custom Models/Infra)
Complex geometry representations and the need to align symbolic CAD constraints with natural-language or high-level design intents will limit scalability; computational cost of running large models in a tight design loop is another constraint.
Early Adopters
Compared to generic AI design tools or text-to-CAD demos, this work focuses specifically on an agentic, human-in-the-loop workflow for industrial geometric modelling, which better matches the safety, accuracy, and accountability requirements of construction and industrial design practices.