This work explores the potential of a large-language-model (LLM)-driven chatbot training tool to mitigate PSA, addressing the common negative perception of public-speaking skills among students. While most existing studies focus on anxiety during scripted presentations, this formative diary study examines the effectiveness of two versions of a chatbot application (virtual reality (VR) and desktop) in reducing anxiety during question-and-answer (Q\&A) sessions. Our study aims to assess the impact of the chatbot training tool on graduate students' PSA, explore its practical applications, and compare the effectiveness of both versions against each other. The findings offer valuable insights for developing more effective and sustainable public-speaking training interventions.