Retrieval-Augmented Generative AI for Academic Viva Preparation: Effects on Anxiety, Confidence, and Oral Performance

Abstract

University students need to face high-stakes oral exams - such as academic viva examinations, thesis defenses, proposal presentations and others - that can cause a significant amount of anxiety. This methodological training manuscript illustrates the evaluation procedure of a retrieval-augmented generative artificial intelligence system to prepare a personalized academic viva. The system accesses uploaded academic documents and automatically creates questions that could be asked by an examiner, grades the responses, gives formative feedback to the student, and produces a personal performance summary. A synthetic pre-post dataset was created to demonstrate the proposed analysis for 30 students. In the results, mean academic anxiety decreased from 6.18 (SD = 0.86) to 3.44 (SD = 0.92), t (29) = -23.48, p < .001, Cohen’s dz = -4.29. Mean academic confidence increased from 4.17 (SD = 0.69) to 6.65 (SD = 0.86), t (29) = 19.87, p < .001, dz = 3.63, while rubric-based oral performance increased from 2.95 (SD = 0.46) to 4.07 (SD = 0.47), t (29) = 19.17, p < .001, dz = 3.50. The possibility of 4 themes emerged from the synthetic reflections: increased preparedness, decreased fear of questioning, improved response structure, and ongoing need for human oversight. The findings provided here are not empirical and are meant to be used for data collection, statistical analysis and reporting in APA style.