Publication

Conversational AI meets mindfulness: Exploring LLMs as a socio‑emotional layer in a math intelligent tutoring system

Vera Rief, Mirella Hladký, Minju Yoo, Stephanie Heel, Shintaro Sato, Tomohiro Nagashima

ECTEL2026 · 2026

@inproceedings{rief2026conversational,
  title     = {Conversational AI meets mindfulness: Exploring LLMs as a socio‑emotional layer in a math intelligent tutoring system},
  author    = {Vera Rief and Mirella Hladký and Minju Yoo and Stephanie Heel and Shintaro Sato and Tomohiro Nagashima},
  booktitle = {ECTEL2026},
  year      = {2026},
}

Abstract

Intelligent Tutoring Systems (ITSs) traditionally focus on cognitive support, often overlooking students’ emotional state. We developed an ITS that leverages Large Language Models (LLMs) to provide both cognitive and socio-emotional support in algebra through a pedagogical agent “Matt”. The system offers LLM-based mindful chats, breathing exercises, and mindful hints and feedback based on Kabat-Zinn’s seven principles of mindfulness to support students’ learning experiences and reduce math anxiety. In a classroom study with 7th graders, we compared a Mindful version against Cognitive support-only version. Although external disruptions limited the final analysis to 42 students, the results provide important exploratory insights. While no significant differences were found in math learning or state-math anxiety, the Mindful group reported significantly higher perceived socio-emotional support and warmth from the agent. Exploratory log data revealed that students in the Mindful condition completed fewer problem-solving steps with fewer hints requested. Our study demonstrates the feasibility of integrating mindfulness into ITSs through LLM-based interactions and positions LLMs as an adaptive, socio-emotional layer within cognitive math tutoring.