Diagram illustrating Taisei’s AI-based system that analyzes tender documents, project data and internal knowledge to automatically generate draft construction plans while mitigating information errors.

Drawing on procurement information for public works and the company’s accumulated technical knowledge, the system automatically generates draft documents compliant with the Ministry of Land, Infrastructure, Transport and Tourism (MLIT) format. Even employees with less-experienced can prepare accurate plans quickly, reducing work hours by 85% compared with conventional methods and significantly improving both efficiency and quality.

The system is part of Taisei’s broader initiative to advance the digital transformation of production processes. It employs a multimodal generative AI built on a cutting-edge vision language model (VLM).

By inputting procurement data—such as bid announcements, project outlines, and special specifications—together with Taisei’s internal knowledge base accumulated through past proposals, the AI extracts the information needed for preparing a comprehensive construction plan. Using this data and examples of past company-produced plans, the multimodal AI appropriately combines technical terminology, text, and diagrams to automatically generate a draft plan in about ten minutes.

Taisei has also developed technology to automatically detect sections where VLM output reliability may be low, suppressing misrecognition and misinformation in the resulting file. This mitigates one of the key risks of generative AI—hallucination. By having experienced engineers review only the automatically flagged sections, the company can create high-quality construction plans more efficiently.

The system also supports the transfer of technical expertise. Through the process of drafting construction plans, users naturally learn civil engineering practices, as well as knowledge related to safety and quality management. This enables simultaneous improvement of construction quality and workforce development.