ExpertAgent: Enhancing Personalized Education through Dynamic Planning and Retrieval-Augmented Long-Chain Reasoning

TL;DR

The paper tackles the reliability, personalization, and adaptability gaps in AI-enabled education by introducing ExpertAgent, an interactive learning agent that combines dynamic planning, retrieval-augmented generation (RAG), and long-chain reasoning anchored to a validated curriculum. A continuously updated student model drives real-time instructional planning and targeted feedback, reducing content hallucinations and increasing trust. The authors contribute an integrated architecture that leverages RAG, CoT reasoning, and dynamic planning to deliver proactive, personalized teaching, supported by an internal usability evaluation showing strong acceptance for core functions and usability, with opportunities to improve social adoption. Overall, ExpertAgent demonstrates a scalable path toward trustworthy, adaptive AI tutors capable of enhancing learning efficiency and engagement across diverse subjects.

Abstract

The application of advanced generative artificial intelligence in education is often constrained by the lack of real-time adaptability, personalization, and reliability of the content. To address these challenges, we propose ExpertAgent - an intelligent agent framework designed for personalized education that provides reliable knowledge and enables highly adaptive learning experiences. Therefore, we developed ExpertAgent, an innovative learning agent that provides users with a proactive and personalized learning experience. ExpertAgent dynamic planning of the learning content and strategy based on a continuously updated student model. Therefore, overcoming the limitations of traditional static learning content to provide optimized teaching strategies and learning experience in real time. All instructional content is grounded in a validated curriculum repository, effectively reducing hallucination risks in large language models and improving reliability and trustworthiness.

Figures (5)

• Figure 1: Overview of the ExpertAgent framework.
• Figure 2: Functional flow diagram of the ExpertAgent system interface. The interface consists of three main components: (1) Learning Modules, which allow switching between different learning modes; (2) Select Topic, which enables learners to enter specific study topics; and (3) Smart Chatbot, which provides interactive Q&A and explanations based on RAG retrieval. The overall learning flow follows a sequence of “Module Selection, Topic Selection, AI Q&A,” forming a comprehensive personalized learning pathway.
• Figure 3: Structural diagram of the ExpertAgent teaching interface. The interface consists of six main components: (A) Brief Summary, which provides a quick overview of the selected topic; (B) Content Type Selection, which allows learners to choose different learning formats; (C) Knowledge Details section, which provides structured and hierarchical content including definitions, features, importance, connections, and examples, helping learners gain an in-depth understanding of the topic and build systematic knowledge; (D) AI Q&A area, powered by RAG technology, which enables learners to interact with the intelligent assistant and receive personalized explanations; (E) Retrieved Snippets section, which supplies supporting evidence and contextual references from relevant documents; and (F) Learning Feedback area, which collects learners’ evaluations of the usefulness of the generated content. These components together form a complete loop from knowledge presentation to interactive questioning and feedback-based improvement.
• Figure 4: Practice and feedback interface of the ExpertAgent system. The interface consists of five main components: (A) Practice Session, which includes topic selection and practice arrangements; (B) Answer Review, which displays learners’ correct and incorrect responses with instant feedback; (C) Learning Advice, which provides review recommendations and reference materials based on test results; (D) Knowledge Map, which visualizes the relationships between different learning topics and the learner’s mastery status (blue = Untouched, yellow = Learning, green = Mastered, red = Weak); (E) Feedback Area, which collects learners’ evaluations of the usefulness of the questions; and (F) AI Learning Tips, which generates personalized study guidance and improvement suggestions based on the learner’s errors and weaknesses. Overall, the interface supports a complete learning loop from test completion to result feedback, learning advice, knowledge visualization, and personalized improvement.
• Figure 5: Average scores of the User Acceptance Model across four categories