Every building in the built environment requires extensive documentation that must comply with current legislation. In this work, the possibilities of using large languange models to automate the preparation and completion of building documentation in accordance with Slovenian legislation are investigated. In the thesis we have tested the hypothesis that by using large language models we can prepare adequate construction documentation and automate the process itself. The first part of the thesis deals with the theoretical background of large language models, focussing on the ChatGPT 4o and the construction documentation required fort he constructionof the building. The practcal part of the thesis demonstrates the verificiation of the effectiveness of a large language model in creating and filling out construction documentation forms using pre-built prompting techniques. In carrying out this work, we used the paid version of ChatGPT 4o to automate the creation of construction documentation. The application is multimodal, easy to use and provides transparent results. In this work, based on the tests performed, we found out tha ChatGPT 4o is able to efficiently and correctly compile important parts of the project documentation, especially the database fort he forms describing the title page of the project documentation and the experts involved in the project (Annex 1A, Annex 1B and Annex 2A). However, the results show that the model did not achieve the same level of accuracy and compliance with the regulations when creating the remaining annexes such as Annes 4A, Annex 4B and Annex 4C (forms for obtaining a building permit).
|
