The conservation of historical monuments is a complex, multidisciplinary process encompassing expertise from various fields and relying on diverse information sources. It requires the collaboration of teams of experts addressing a wide range of interconnected challenges. Domain experts are an invaluable source of information due to their immense knowledge and extensive experience with monuments, often with the specific one in question. However, this expertise tends to remain compartmentalized and gets lost over time due to limitations in documentation techniques, interdisciplinarity, and language barriers. To effectively utilize this critical knowledge, improving the existing documentation techniques by incorporating a structured approach to capture knowledge is essential. The potential of AI models, particularly Large Language Models (LLMs) combined with Heritage Building Information Modelling (HBIM), offers significant opportunities for advancement in data collection and processing. However, the capture, structuring, and storage of knowledge related to historic structures with the capabilities of AI remains a widely underexplored area. The approach presented addresses these challenges by focusing on the Lausanne Cathedral, Switzerland, as a case study. Information from diverse sources, such as voice recordings, literature (i.e., books, reports, archives), and on-site observations, is converted into a comprehensive database containing details on material properties, age, and degradation using an LLM. The database is mapped onto the monument?s HBIM model via retrieval augmented generation (RAG) and a Python API. This paper introduces a workflow for AI-driven documentation of knowledge primarily transmitted orally and supported by other mediums.