Performance Comparison and Improvement of LLM Models for Automated Generation of Fire Protection System Construction Specifications

This study verifies the effectiveness of using major LLMs (Gemini, Claude, ChatGPT) to automatically generate construction specifications for fire protection systems in specific fire-service-designated objects like factories, warehouses, and hospitals, and conducts a comparative analysis of each model's performance. The evaluation shows that while LLMs generate the basic structure and general content of the specifications well, they have limitations in reflecting the unique characteristics of the target buildings and accurately applying the latest laws. For complex facilities such as hospitals, sophisticated prompt design and expert review were essential. This research confirms the potential of LLMs for document automation in the fire safety field and provides foundational data for the future development of specialized language models for fire protection.