Generative AI for Military Facility Construction Administration

Overview

The Ministry of National Defense (MND) has faced frequent legal disputes and accident risks due to unauthorized or unregistered buildings within military bases. Coupled with the professional limitations of administrative staff and a surge in civil complaints, a systematic solution became essential. To resolve this, the MND developed the "Architectural Regulation Copilot," which integrates scattered laws, administrative rules, ordinances, and design standards. By providing user-centric, easy-to-understand legal guidance to various stakeholders, this service aims to eliminate confusion during the permit and construction process while enhancing the efficiency of architectural administration.

Project description

To optimize AI model training, the current legal-centric architectural regulations were restructured into a user-centric classification system—categorized by administrative rules, technical standards, and building lifecycle stages. Data is managed at the specific level of articles and clauses to enhance embedding accuracy. Furthermore, unstructured civil complaint data, including records from the e-People (Kookmin Shinmun-go) portal, was processed for supplementary training. The core development includes three specialized modules: "Architectural Admin AI" for permit support, "Architectural Design AI" for design and construction tasks, and "Architectural Complaint AI" for public assistance.

Through this Copilot service, permit officers receive administrative support, practitioners obtain assistance in design, construction, and supervision, and general users benefit from streamlined civil complaint responses.

The system is built upon a verified "Smart DB" from the Korea Building Regulation system, allowing users to view laws, enforcement decrees, and regulations at a glance and stay updated on frequent revisions.

Unlike traditional post-processing methods—which often resulted in fines or enforcement levies due to overlooked regulations—this service provides a preventative legal review at every stage of the construction process, enabling both institutions and citizens to verify compliance in real-time.

Key technological features

• Specialized Infrastructure: Construction of a Smart Database for architectural regulations and a domain-specific sLLM (small Large Language Model)
• Advanced RAG Integration: Implementation of a dedicated RAG (Retrieval-Augmented Generation) framework to seamlessly combine the legal database with the sLLM

Expected outcomes

The development of this Generative AI-based service efficiently organizes complex architectural laws and provides regulation services through a user-friendly classification system. This allows for easier identification of restrictions and improved management systems, preventing operational confusion even when non-experts are involved in administrative procedures. In particular, the ability to verify regulatory compliance across the entire lifecycle—from design to permit approval—is expected to drastically reduce administrative lead times.

The MND anticipates significant quantitative improvements, including reducing the administrative cancellation rate from over 10% to below 5%, increasing budget execution rates by more than 3%, and shortening processing times by over 30%. Local governments are also expected to see administrative burdens lightened, with processing and complaint response times reduced by 20% and 30%, respectively. By providing diverse services including legal interpretations, precedents, and Q&A, the system has bolstered trust between public institutions and private users. Aligned with the Defense Innovation 4.0 plan, this Copilot service will support highly specialized new projects, ensuring a safe environment through strict regulatory compliance and enhancing public trust in military facilities.

Regarding the expansion plan for this service, a directive has been issued to the Army, Navy, Air Force, Marine Corps, and units directly under the Ministry of National Defense (MND)—which are responsible for facility project requirements exceeding 5 trillion KRW annually—as well as to intermediate review authorities. These entities are instructed to utilize the system and submit improvement requests. The consolidation of these key issues and feedback is scheduled to be completed within the first half of 2025.

To further advance the service, the MND is currently securing budget and establishing operational strategies in connection with the "Next-Generation Integrated Defense Facility Information System Project," which is set to launch in 2025. This integration will ensure the long-term sustainability and technological sophistication of the Architectural Regulation Copilot within the broader defense infrastructure.

Launch year: 2024

Attribution

Case study courtesy of the National Information Society Agency, Republic of Korea, Apolitical's Content Partner. Sourced from Use Cases of Public AI Service, Vol.2, https://eng.nia.or.kr/site/nia_eng/ex/bbs/View.do?cbIdx=31975&bcIdx=29182&parentSeq=29225