Educational programs emphasizing experiential learning and personal development increasingly rely on student reflections to assess learning outcomes beyond traditional grade-based measures. However, manual analysis of open-ended textual reflections is time-consuming, subjective, and particularly challenging for short texts. This paper applies an automated topic detection method using sentence embeddings and unsupervised machine learning to analyze student reflections and identify learning gains expressed in students’ own words. Our approach combines sentence embedding techniques to capture semantic information from short textual reflections with K-means clustering to automatically partition embeddings into meaningful topic clusters. This methodology enables systematic identification of themes that students emphasize when describing their learning experiences, providing insights into personalized learning outcomes that traditional assessment methods might miss. We demonstrate this approach using a corpus of short reflections from service-learning programs. The method is broadly applicable to any educational context where understanding student experiences and feelings takes precedence over conventional assessment metrics. Experimental results show the method effectively uncovers meaningful topics from brief, open-ended reflections. Analysis of frequent keywords within each topic cluster provides fine-grained insights into student perspectives on their learning outcomes. Our findings reveal that students with higher self-reported learning outcomes tend to discuss a broader range of topics in their reflections. This computational approach offers educators and researchers a scalable tool for understanding what aspects of educational experiences most significantly impact student learning from the learners’ own perspectives, with applications extending beyond service-learning to diverse experiential education programs.