AEGIS-CAD

Agentic Engineering with Geometric Intelligence and Simulation for Automated CAD Generation. The system takes heterogeneous engineering inputs — 2D pipe drawings, existing 3D assemblies, technical documentation, and natural-language requirements — and builds from them an editable parametric 3D model of the route, fitted into a constrained volume while respecting connection points, tolerances, and assembly limits. In parallel it prepares a structural analysis: mesh generation, refinement strategy, convergence analysis, and detection of critical stress zones; the final decision is always approved by an engineer. The key idea is compactness through verified analysis: a layout can be tightened exactly as far as a converged strength and modal analysis confirms, with verification built into the generation loop as an admissibility criterion. The system is not tied to a specific CAD package — agents work with a software-independent engineering model, and adapters translate verified specifications into commands of a particular CAD/CAE system through its API (the reference stack is CadQuery/OCCT, FreeCAD, Gmsh, CalculiX; the first target commercial adapter is SolidWorks). Status — research prototype AgentCAD v3 (TRL 3), target level TRL 5. The team is six ZNU researchers (computational mechanics and FEM, multiphysics modeling, engineering software, machine learning), led by Assoc. Prof. Oleksii Kudin, PhD. Developed for the AI-BOOST AI Challenge Competition (Horizon Europe), the challenge "Agentic AI for Automated CAD Generation and Autonomous Simulation" posed by the industrial group SIAD.