Personalized E-Learning Assistant Using Knowledge Graph and LangGraph-Orchestrated Multi-Agent Framework

Abstract

The design of an e-learning system incorporating a personalized instructional approach, surpassing conventional pedagogical methodologies, is presented in this study. Many adaptive tutorial systems don\'t consider students\' existing knowledge and past learning well enough. To fix this, the system uses knowledge graphs and multi-agent systems. The main focus is to build an academic system that can record students\' likes learning styles, and grades in a clear way. A semantic graph model has been set up in Neo4j to store and manage this data. The system uses modular agents run by Lang Graph, to work with the knowledge graph. This allows it to change learning paths, teaching content, and make extra materials that fit each student. As students use the system more and take quizzes, each round of learning gets better to stay relevant and teach well. Adding Django makes the system more flexible and able to work with different front-end apps. This approach stands out because it\'s forward-thinking, can change, and aims to make learning better over time making it a strong choice compared to other options.

Introduction

Limitations of Current E-Learning Platforms:
Most existing e-learning systems use fixed, linear curricula that fail to accommodate the diverse backgrounds, learning speeds, styles, and needs of individual students. Attempts at personalization often focus on superficial preferences rather than real learning behavior, resulting in less engagement and poorer learning outcomes.

Proposed Personalized Approach:
A truly adaptive e-learning system should dynamically adjust to each learner’s interactions and progress, not follow a preset path. The suggested solution uses a knowledge graph to represent each student’s learning state (strengths, challenges, goals) and an agent-based architecture to personalize content delivery and assessment in real time. Large language models (LLMs) generate tailored explanations, quizzes, and learning plans aligned with the learner’s preferences and progress.

Literature Insights:
Research shows knowledge graphs effectively model complex subject dependencies and learner states, enabling tailored educational pathways. Agent systems combined with LLMs enhance personalized content generation and feedback loops, making learning adaptive and engaging. Retrieval-augmented generation (RAG) and multi-agent architectures further improve response relevance and interactivity.

Proposed System Architecture:

• A Neo4j-based knowledge graph stores detailed learner profiles, academic topics, resources, and quiz data.

• Multiple agents orchestrated by LangGraph manage user interaction, query the graph, plan learning paths, and generate quizzes.

• LLMs (e.g., GPT, Gemini) create personalized content and evaluate answers, dynamically adjusting prompts based on learner data.

• APIs enable communication between frontend and backend, maintaining modularity and scalability.

Methodology:

• Onboarding gathers comprehensive student data for graph population.

• Agents use LangGraph for request routing and task management.

• Quizzes are dynamically generated and scored, with results updating the knowledge graph.

• Learning plans adapt continuously based on quiz outcomes and evolving learner profiles.

Results:
The personalized system outperforms traditional e-learning platforms and static LLM-based methods in quiz relevance, personalization, and adaptability. It provides tailored learning paths, intelligent quiz generation, and continuous adjustment, resulting in a more engaging, effective learning experience.

Conclusion

The presented system demonstrates the efficiency of combining structured knowledge representation through a semantic knowledge graph with modular agent-based orchestration to achieve deep personalization in e-learning environments. The proposed approach extends beyond traditional content delivery by continuously adapting learning pathways based on student profiles, preferences, performance, and contextual feedback. By enabling dynamic generation and refinement of learning resources, the system addresses key limitations found in conventional educational platforms, particularly in terms of engagement, relevance, and learner autonomy. The incorporation of participatory and human-centred design principles further ensures that the framework remains responsive and inclusive, reversing the one-size-fits-all paradigm commonly observed in legacy systems. Future enhancements are envisioned to include interoperability with external educational platforms such as MOOCs and content-sharing networks, the development of real-time progress tracking interfaces for mentors and guardians, and the inclusion of behavioral and affective signals for enriched learner modeling. Through these developments, a step forward is taken toward realizing intelligent tutoring systems that are seamlessly personalized, minimally intrusive, and capable of supporting diverse learning needs at scale.

References

[1] R. Kumar and S. Patel, “LangGraph-Orchestrated Multi-Agent Systems for Personalized Course Recommendations,” unpublished manuscript. [2] S. Bhuvanesh et al., “Knowledge Graph Based Medical Chatbot,” in Proc. IEEE Global Conference for Advancement in Technology (GCAT), 2023. [3] H. Aberbach et al., “A Personalized Learning Approach based on Learning Speed,” Journal of Computer Science, vol. 17, no. 2, pp. 112–120, 2021. [4] S. Zhao et al., “Advancements and Transformations in Educational System Reforms,” SHS Web of Conferences, vol. 165, 2024. [5] LangChain Documentation. [Online]. Available: https://docs.langchain.com [6] LangGraph GitHub Repository. [Online]. Available: https://github.com/langchain-ai/langgraph [7] S. Zerhoudi and M. Granitzer, “PersonaRAG: Enhancing Retrieval-Augmented Generation Systems with User-Centric Agents,” arXiv preprint arXiv:2407.09394, 2024. [8] Y. Shi et al., “ERAGent: Enhancing Retrieval-Augmented Language Models with Improved Accuracy, Efficiency, and Personalization,” arXiv preprint arXiv:2405.06683, 2024. [9] M. Thway et al., “Battling Botpoop: RAG Chatbot for Higher Education,” arXiv preprint arXiv:2406.07796, 2024. [10] D. Zhou et al., “Evaluating Knowledge Graph Based Retrieval-Augmented Generation Methods under Knowledge Incompleteness,” arXiv preprint arXiv:2504.05163, 2024. [11] N. Niyozov et al., “The Pedagogical Principles and Effectiveness of Utilizing ChatGPT for Language Learning,” E3S Web of Conferences, vol. 385, 2023. [12] A. Bijanov et al., “Blended Learning and Personalized Feedback in Modern Education,” Proc. RSES Conference, 2023. [13] H. Wan, B. Che, H. Luo, and X. Luo, “Learning Path Recommendation Based on Knowledge Tracing and Reinforcement Learning,” unpublished manuscript. [14] Y. Zhang et al., “Dynamic Knowledge Graph Construction for Adaptive E-Learning Systems,” unpublished manuscript. [15] A. Joshi et al., “Scalable Multi-Agent Frameworks for Real-Time Learning Analytics,” unpublished manuscript. [16] P. Gupta et al., “TAGCN-Based Knowledge Graphs for Learner Preference Modeling,” unpublished manuscript. [17] X. Wang et al., “Federated Knowledge Graphs for Privacy-Preserving E-Learning,” unpublished manuscript. [18] Z. Li et al., “CrewAI-Driven Task Decomposition for Collaborative Learning Agents,” unpublished manuscript. [19] M. Singh et al., “Quantum-Inspired Optimization for Knowledge Graph-Based Tutoring Systems,” unpublished manuscript. [20] D. Fernandez et al., “Streaming Knowledge Graphs for Real-Time Learner Feedback,” unpublished manuscript.

Copyright

Copyright © 2025 Chitresh Gyanani, Owais Salauddin Khan, Arjun Singh, Sameer Asthana. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.