The ability to share data can facilitate cooperation and decision making throughout the entire life cycle of buildings, from initial stages of planning, through design and construction, up to the management of assets and towards end of life and recycling or reuse. Accordingly, the different actors of the Construction industry can share data and functionalities across software platforms through automated processes. Such processes involve stakeholders with heterogeneous backgrounds; for this reason, it is of value to make data available to people without expert knowledge of specific programs or computer systems.

This study is concerned with digital protocols for automated capturing of real-time sensor data for the assessment of building performance. The research explores a Large Language Model (LLM) driven protocol for sensor data acquisition to evaluate building performance, leveraging the existing database where building data is stored. Using pre-written prompts which instruct how to carry out tasks in a step-by-step manner, the LLM can utilize pre-defined functions built upon API calls for communication with sensor data, including data acquisition, post-processing, and interpretation. The LLM receives instructions by the User using natural language. Initial tests underscore the protocol feasibility, highlighting its potential utility to improve the data communication of existing digital twins for individuals without professional expertise in building management. A case study exemplifies how human instructions in natural language trigger the LLM to invoke a request for indoor climate sensor data in a building.

The developed tool was tested by professionals working in the Construction industry and the Educational Sector to provide feedback on its practical application. This evaluation helped to identify weaknesses for future developments and proved the flexibility and adaptability of the method.

As advancements in Artificial Intelligence (AI), particularly in LLMs, continue to surge, there is anticipation for further refinements. This includes cost reduction, enhanced stability, and the integration of more advanced functionalities such as advanced data analysis using machine learning coded by LLM.