Da Vinci: Recursive Improvement CAD Harness

Da Vinci is a recursive improvement CAD harness designed to support physical engineering tasks through specialized AI agents, programmatic CAD and automated evaluation. Powered by GPT-6 Astra, agents generate designs, evaluate results, create and test reusable Python tools, and carry lessons into subsequent iterations. MongoDB Atlas stores designs, evaluations, tools and policies; Vector Search retrieves relevant experience, GridFS preserves CAD artifacts, and Database Triggers initiate evaluation and reflection workflows. In a drone sensor-mount demonstration, the harness reduced mass from 83.5 g to 30.3 while meeting fixed stress and deflection screening limits. A Next.js interface presents every design as an interactive 3D model with quantitative comparisons. The architecture can extend to other physical engineering problems by integrating task-specific geometry, constraints and evaluators.