Students often struggle to reason effectively through formal logic problems. This project explores how strategic and reflective thinking emerge during problem solving in Deep Thought, an intelligent tutoring system for propositional logic. Using multiple data modalities–student conversation and system interaction logs–we aim to understand how different forms of reasoning contribute to performance and learning. We apply thematic analysis to student think-aloud sessions to identify segments of reasoning behavior, such as goal formation or self-correction. These tags inform a qualitative coding scheme used to analyze reasoning patterns across participants. In parallel, we use natural language processing (NLP) techniques to detect similar reasoning behaviors directly from raw conversation transcripts, enabling scalable classification of dialogue features. Finally, Python-based analysis of the numerical interaction logs allows us to examine how behavior unfolds over time in relation to hint usage, efficiency, and task completion. Together, these methods provide a dynamic view of student reasoning and lay the groundwork for real-time metacognitive support in education software.
