Collaborative Multi-Agent Chatbot Framework for Intelligent Tutoring and Feedback
No Thumbnail Available
Date
2026
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
SDU University
Abstract
Intelligent Tutoring Systems (ITS) seek to deliver a personalized approach akin to one-on-one human tutoring. The advent of large language models (LLMs) makes it possible to create conversational tutoring systems at a reasonable scale. Nevertheless, despite some recent advances, single-agent LLM-based tutoring still suffers from problems such as inaccuracies, varying quality of the pedagogical content, and low self-correction ability. In this dissertation, a Collaborative Multi-Agent Chatbot Framework (CMACF) for ITS and assessments is presented and evaluated. The system includes three agents: Tutor Agent, which generates adaptive explanations; Feedback Agent, which detects and corrects any potential misunderstandings; and Verifier Agent, which validates and optimizes responses prior to their delivery to the learner during Java programming lessons. For testing, the framework was used to perform 40 different Java tasks in the following domains: loops, conditionals, arrays, object-oriented programming, exceptions, and typecasting. A comparison between three frameworks was conducted based on three scenarios: single-agent system (SA), multi-agent system without verification (MA-NV), and the proposed multi-agentsystem with verification (MAV). The criteria applied included correctness, clarity, pedagogy, actionability, and consistency of responses. It can be concluded that the proposed model resulted in the best scores (4.52 points) and surpassed both the single-agent solution (4.165) and the multi-agent solution without verification (4.215). In addition, 25% of generated responses were subject to changes due to the verification component and improved task completion rate from 17.5% to 52.5%. Although the response generation time of MA-V (11.1 seconds) exceeded the single-agent solution (3.66 seconds), its quality is sufficient for intelligent tutoring purposes.
Description
Keywords
Intelligent Tutoring Systems, Multi-Agent Systems, Large Language Models, Formative Feedback
Citation
Myrzaliyeva Zh / Collaborative Multi-Agent Chatbot Framework for Intelligent Tutoring and Feedback / SDU University / Department of Computer Science