Customization and personalization of large language models for engineering design
Zhoumingju Jiang, Ang Liu, Dawen Zhang, Xiwei Xu, Yun Dai
Abstract
Large language models (LLMs) are increasingly used in design and manufacturing, yet directly employing general-purpose LLMs for conceptual design often leads to unmanufacturable concepts. This paper aims to adapt general-purpose LLMs for design-specific tasks. A new framework is presented to customize a general-purpose LLM into a design-specific model based on design-relevant data and Retrieval-Augmented Generation (RAG). Another complementary framework is presented to personalize the design-specific LLM by integrating design reasoning with prompting techniques. A design experiment, using patent documents as the design-relevant data, demonstrates that customization and personalization can improve LLM effectiveness in conceptual design, especially by enhancing concept feasibility.