CADDesigner: Conceptual Design of CAD Models Based on General-Purpose Agent

Computer Aided Design (CAD) plays a pivotal role in industrial manufacturing but typically requires a high level of expertise from designers. To lower the entry barrier and improve design efficiency, we present an agent for CAD conceptual design powered by large language models (LLMs). The agent accepts both textual descriptions and sketches as input, engaging in interactive dialogue with users to refine and clarify design requirements through comprehensive requirement analysis. Built upon a novel Explicit Context Imperative Paradigm (ECIP), the agent generates high-quality CAD modeling code. During the generation process, the agent incorporates iterative visual feedback to improve model quality. Generated design cases are stored in a structured knowledge base, enabling continuous improvement of the agent’s code generation capabilities. Experimental results demonstrate that our method achieves state-of-the-art performance in CAD code generation.

💡 Research Summary

The paper introduces “CADDesigner,” a groundbreaking AI agent powered by Large Language Models (LLMs) designed to revolutionize the conceptual design phase of Computer-Aided Design (CAD). Traditionally, CAD modeling has been a high-barrier task requiring specialized expertise and significant time investment. CADDesigner aims to democratize this process by enabling intuitive, automated, and interactive design workflows.

The architecture of CADDesigner is built upon several sophisticated technical pillars. First, it features a multimodal input system that accepts both textual descriptions and hand-drawn sketches. This allows the agent to capture a rich spectrum of design intent, ranging from high-level functional requirements to specific geometric nuances. To ensure accuracy, the agent engages in an interactive dialogue with the user, performing a comprehensive requirement analysis to clarify ambiguities before the generation process begins.

A core innovation presented in this research is the “Explicit Context Imperative Paradigm (ECIP).” Unlike standard LLM prompting, which can struggle with the precise geometric constraints required for CAD, ECIP provides the model with an explicit framework of context and imperative instructions. This paradigm ensures that the generated modeling code adheres strictly to the necessary mathematical and structural constraints of 3D modeling, resulting in high-fidelity CAD outputs.

Furthermore, the agent implements an “Iterative Visual Feedback” mechanism. This creates a closed-loop system where the generated code is rendered into a 3D visual representation, which the agent then “inspects” as a visual input. By comparing the rendered output against the original design intent, the agent can autonomously identify discrepancies and iteratively refine the code. This self-correcting capability significantly enhances the robustness and quality of the final design.

To ensure long-term intelligence and continuous improvement, CADDesigner utilizes a structured knowledge base. Every successful design case is archived in a structured format, allowing the agent to retrieve and learn from past successes. This creates a flywheel effect where the agent’s code generation capabilities are constantly upgraded through accumulated experience. Experimental results demonstrate that CADDesigner achieves state-of-the-art performance in CAD code generation tasks, marking a significant leap toward fully autonomous, AI-driven engineering design.