Danry, V*., Billa, J. G.*, Samaradivakara, Y.*, Liang, P. P., Maes, P. (2026, March). Mind Mapper: Modeling and Predicting Behavioral Patterns from Everyday Conversations with Wearable AI Systems and LLMs. In Proceedings of the 31th International Conference on Intelligent User Interfaces.

Abstract

Everyday conversations are more than exchanges of words—they reveal how people think, react, and adapt across situations. Through our speech we reveal the cognitive patterns that shape our behavior over time. Yet much of these patterns remains implicit: people operate through recurring heuristics and habits—some helpful, some limiting, many unnoticed. However, identifying such patterns requires self awareness that humans struggle with—and that current user modeling systems, often fragmented and narrowly tailored to specific tasks, fail to capture. Recognizing these patterns can enable user support systems to go beyond reactive assistance towards anticipatory support. We present Mind Mapper, an always-on wearable AI system that mines behavioral patterns from everyday real-life conversations. Mind Mapper employs a multi-stage LLM pipeline to generate, refine, and evaluate human-readable behavioral patterns. In a field study with 12 participants capturing over 700 hours of real-life conversational data, Mind Mapper generated behavioral patterns that participants consistently rated as accurate, unique, and helpful for reflection and behavior change. We further illustrate how such behavioral pattern modeling might enable new forms of human-AI interaction—anticipating user behavior through proactive interventions, adaptive content delivery, cognitive reframing, and behavioral simulation. Our results show the potential of always-on wearable systems with LLM-driven user models to support new forms of cognitively scaffolding, context-aware human-AI interactions.