Abstract: In view of the semi-theoretical and semi-empirical characteristics of deep foundation pit engineering design, an intelligent foundation pit design process based on large-language models (LLMs) is proposed. Based on the Grasshopper visual programming plug-in, LLMs are used to assist in writing Python scripts to realize the parametric processing and automatic execution of foundation pit plane design work. This process takes the corner point coordinate positioning of the foundation pit as the input end and can output relevant results of the retaining structure, support system and design drawing. The research results have been verified by engineering cases. The design cycle has been compressed from the traditional 3-4 days to 20-30 minutes, which significantly improves the design efficiency and provides technical guarantee for the rapid response of complex foundation pit projects. Research shows that it is feasible for LLMs to empower deep foundation pit engineering design, and it also has certain engineering practical value, providing new ideas for the digital and intelligent transformation of the industry, and laying a certain foundation for the in-depth application of artificial intelligence technology in the field of civil engineering in the future.