The paper describes the development of a methodology for structured recording and searching of the engineering process during product design using advanced tools based on large language models (LLM). In the introductory part of the work, various existing approaches to recording design processes were investigated, from analogue methods to modern digital solutions such as DRed (Design Rationale Editor) and IBIS diagrams. The applicability of large language models for the analysis and interpretation of notes and screenshots created during engineering work is investigated. For this purpose, different versions of the commercially available ChatGPT large language model (o1, o3-mini, o3-mini-high, 4o-mini, 4o and 4.5) were tested in detail. Experimental testing has clearly demonstrated that new versions of these models provide significantly more accurate results in terms of understanding design requirements, technical specifications, identifying key changes, and creating structured notes. Through a series of experimental tests of the methodology, including the tasks of choosing a bearing, designing an electric motor board and modelling an LED lamp, it was determined that the application of LLM significantly contributes to the quality and efficiency of recording the construction process. In particular, the automatic generation of the engineering log and IBIS diagrams based on the collected data has proven to be useful, ensuring transparency, ease of access to information and improved knowledge management within the team. The results show that more advanced versions of the ChatGPT model provide more precise analyses and more detailed insights, while older versions have significant limitations in the interpretation of specific technical information. In conclusion, the integration of large language models into engineering note-taking processes has proven to be a promising practice with great potential for further improvement.
Tibor Totman
2025
Master thesis
