The diploma thesis examines the automation of the preparation and processing of construction documentation using large language models (LLMs). It addresses the problem of extensive, dispersed and often unstructured project documentation, where manual searching, verification and preparation of content require considerable time and increase the possibility of errors. The aim of the thesis was to assess in which procedures LLMs can effectively support work with construction documentation and where, due to technical and legal responsibility, expert human judgement remains necessary. The theoretical part presents the characteristics of construction documentation, the basics of artificial intelligence, and the differences between rule-based processing of structured data and probabilistic data processing with LLMs. The practical part is based on project documentation for construction execution, on which procedures for data extraction, content analysis, compliance checking and generation of draft documents were tested. The results show that LLMs are highly reliable when dealing with directly stated data. However, in tasks that require linking several sources, excluding similar but irrelevant data, or generating technical text, deviations, incorrect interpretations and generalisations may occur. It was established that LLMs can significantly accelerate the review of documentation, reduce the amount of manual work and assist in the preparation of drafts; however, they cannot replace responsible engineering verification. Their meaningful use therefore depends on clearly defined input data, traceable results, data protection and mandatory expert review before practical application.
|
