Adaptive Multi-Agent Tutoring AI for Multimodal Mathematics Conversational Learning
Quitt Le, Hoang D. Nguyen · 2025
Providing tailored and multimodal academic support in mathematics remains a complex task, especially when the goal is to deliver timely assistance aligned with specific curricula. Many current question-answering (QA) systems are limited in interactivity, perform inconsistently across different input types (e.g., text, image, voice), and often fail to respond effectively to individual learning needs. These challenges are even more prominent in subjects like algebra or geometry, where abstract reasoning and step-by-step solutions are crucial. This paper introduces a conversational AI tutoring system grounded in Large Language Models (LLMs) and multi-agent design, with each agent addressing a specific instructional function (Informer, Verifier, Insight, Practice, Tutor). The system integrates multimodal inputs and features a personalized feedback loop that tracks student difficulties, updates learning records, and suggests suitable practice tasks or learning videos. Evaluated with ninth-grade students in Vietnam, the system has demonstrated promising results for both educational impact and broader applicability.