A Scalable LLM-Based Framework for Intelligent and Sustainable Construction Operations

Abstract

The increasing complexity of construction operations requires intelligent computational systems capable of integrating heterogeneous information, supporting operational decisions, and maintaining scalability across projects of different sizes and characteristics. This research develops a conceptual Scalable LLM-Based Framework for Intelligent and Sustainable Construction Operations by synthesizing principles from artificial intelligence safety, intelligent-agent environments, reinforcement learning, human intervention, and evolutionary computation. Because the supplied literature does not directly address large language models (LLMs) or construction management, the proposed framework positions LLM-based construction intelligence as an interdisciplinary extension of established AI-agent and safety principles rather than as an empirically validated construction system. The methodology employs a reference-driven conceptual synthesis in which construction information is represented as a structured operational environment, LLMs function as reasoning and coordination components, and human supervision operates as a safety and governance layer. The framework is organized around five functional dimensions: information integration, intelligent reasoning, adaptive decision support, human-in-the-loop control, and sustainability-oriented operational optimization. The analysis indicates that scalability depends not only on computational capacity but also on the ability to constrain decisions, manage uncertainty, preserve human oversight, and transfer intelligence across heterogeneous operational environments. The proposed framework consequently emphasizes controlled autonomy rather than unrestricted automation. Its principal contribution is a theoretically grounded architecture for connecting LLM-enabled reasoning with safe, adaptive, and sustainable construction operations while explicitly recognizing the limitations of the available evidence.