Conference Session

Industry and Technical Engineering Education

Collection

2025 ASEE Annual Conference & Exposition

Authors

Aakash Alpesh Patel, University of Georgia; Taiwo Raphael Feyijimi, University of Georgia; Sarah Jane Bork, University of Georgia

Tagged Divisions

Student Division (STDT)

Excel file. The retrieved transcripts were thenprocessed to convert them into text from transcript form. This involved the removal of timestamps and correction of word spacing. Stage 3: Transcript Evaluation: For this study, we built off ongoing work by members ofthe research team to adapt a framework to perform deductive thematic analyses [redacted; underreview]. This method leverages a combination of prompt engineering techniques (PETs), naturallanguage processing via large language models (NPL via LLMs; i.e., ChatGPT), and Bradley etal.’s framework on thematic analysis. Appendix B details the exact prompts used to extractrelevant themes and ideas from the transcripts. Bradley et al.’s study outlined a method whereseveral codes should

Conference Session

Engineering Education Methods and Reflections

Collection

2025 ASEE Annual Conference & Exposition

Authors

Jad El Harake, Vanderbilt University; gina yu, Vanderbilt University; Kaden Jorge Tro; Jonathan Ehrman, Department of Biomedical Engineering, Vanderbilt University, Nashville, TN, USA

Tagged Topics

Diversity

Tagged Divisions

Student Division (STDT)

perceived growth and development of the student. In the latter case, manualcoding of the responses revealed which specific skills were acquired by the student and identifiedby the mentor but not by the student response, leading to a positive score discrepancy, or theareas which mentors identified as having room for improvement, leading to a negative scorediscrepancy.When considering the thematic content of all responses rather than focusing on those whichpresented with score discrepancies, coding and tallying of responses was complemented with theaid of the LLM ChatGPT (OpenAI, CA, USA). The use of LLMs in content analysis has beenpreviously shown to have good agreement with human results [12], [13]. In this study, ChatGPTwas prompted to identify