A Conceptual Framework for LLM-based Multi-agent System in Construction Management

Abstract: This research explores LLM-based multi-agent systems for construction management challenges. A framework is proposed, using specialized LLM agents coordinated centrally to automate and optimize project aspects. It integrates diverse memory, APIs, and environmental awareness. Leveraging LLM reasoning and natural language, it incorporates construction knowledge via databases and knowledge graphs. The system demonstrates potential for improved efficiency in scheduling, resource allocation, safety, and decision support. Data quality, computation, and explainability are acknowledged limitations. This work provides a theoretical basis for LLM multi-agent applications in construction.