The aim of this paper is to explore how to take advantage of the Large Language Models’ (LLM) knowledge by integrating the environment description provided by the deployed IoT sensing constellation guided by a BIM interface. By reaching the specific items of interest and the proper context, it is expected a more specific and personalized advise than just the current general rule based systems. The paper will implement a Design Science methodology. The Design principles followed in this research were influenced by the work of 1 and 2. Their design principles involve a structured process of identifying problems, defining objectives, developing and evaluating potential solutions, and finally communicating the results. This process helps to ensure that the resulting artifacts are effective, useful, and relevant to the needs of the stakeholders in a specific context. It is expected to validate the approach as a tool to be taken into account during on-site planning interventions, both during construction but also during operation. The dynamic behavior implemented on both ends, first by setting the focus on specific items, but also by providing detailed and dynamic context can break the general application of rule-based criteria for something more specific and close to the effective conditions. LLMs gained momentum recently but they lack of application and limitation analysis of functionality for real use cases. Until now very few contributions have taken advantage of this idea 3 and 4. This paper will explore to what end using prompting engineering and dynamic context can help in providing accurate and personalized recommendations. This work will exclude improvements coming from the fine tuning of the LLM and will run under a limited context of less than 8000 tokens for the selected models.
Loreiro et al. (Mon,) studied this question.