The 2026 ASME-CIE Student Hackathon wrapped up in Houston, USA. The CoDeFab team from the Southern University of Science and Technology (SUSTech) won third place in the Autodesk challenge with their “Design Intent-Aware Agentic CAD Editing Framework.” This is also the first domestic university team to ever receive this award. 

The competition is organized by the Computers and Information in Engineering Division of the American Society of Mechanical Engineers (ASME-CIE) and serves as a pre-event for the ASME International Design Engineering Technical Conferences & Computers and Information in Engineering Conference (IDETC/CIE). It has been held for six consecutive years. This year’s theme was AI for integrated design-to-manufacturing, focusing on cutting-edge applications of large models in design optimization, CAD model editing, and automated manufacturing planning. The event took place from August 13 to 23 in a hybrid online and offline format, attracting 46 participants from world-renowned universities like MIT, ETH Zurich, McGill University, and the University of Texas at Austin. 

The CoDeFab team consists of Zhoutao BI, a master’s student, and Zhoumingju JIANG, a Ph.D. student, both from the School of Automation and Intelligent Manufacturing, with Associate Professor Yi XIONG as their advisor.

Figure 1. CoDeFab Team’s Competition Entry Page

The Autodesk competition focused on the use of Vision-Language Models (VLMs) for CAD model editing and iterative refinement in real engineering scenarios. The goal was to explore how to accurately and efficiently make intelligent edits to existing CAD models based on design requirements. To address the difficulty VLMs have in accurately understanding design needs and editing CAD models, the team developed a design‑intent‑aware agentic CAD editing framework that senses design intentions. This framework combines natural-language editing requests with visual expectations to build a structured design intent. The system first analyzes user needs and uses image generation models to predict the desired edits. Then, by comparing the existing CAD model with the expected result, it provides a visual basis for inferring design intent. On this foundation, the team developed a closed-loop system based on harness engineering, allowing the agent to dynamically adjust CAD operations based on visual and geometric feedback and, if necessary, re-infer design intent, achieving automatic editing and iterative optimization of CAD models. This framework not only improves CAD editing quality but also reduces token consumption during large model inference.

The competition used a two-round review process. In the first round, teams submitted their code, CAD result files, and a demo video. Teams that advanced moved on to the finals. In the final round, each team gave a 15-minute presentation covering research questions, technical solutions, experimental results, and a Q&A session. After expert evaluation, the CoDeFab team ultimately won third place in the Autodesk competition.

Figure 2. CoDeFab Team at the Finals Award Ceremony

This award fully showcases the innovative strength of SUSTech students at the intersection of artificial intelligence, computer-aided design, and intelligent manufacturing, and also provides new research ideas for applying Multimodal Large Language Models to the field of engineering design.