Personalized AI Tutor Using Language Models and Learning Analytics

Abstract

This research proposes a comprehensive AI tutoring framework that leverages the capabilities of large language models (LLMs) and learning analytics to create a personalized and adaptive learning environment. Unlike traditional Intelligent Tutoring Systems (ITS), which are constrained by static rules and predefined flows, the proposed framework offers dynamic content generation, contextual responsiveness, and real-time performance monitoring. It integrates generative AI technologies with microservice-based architecture to enhance the learning experience through quizzes, feedback, and lesson adaptation. The system is built using modern web technologies such as FastAPI for the backend and React for frontend interaction. A four-week experimental evaluation with undergraduate students indicated significant improvements in engagement, knowledge retention, and learner satisfaction. This study demonstrates that generative AI, when combined with analytics and ethical safeguards, can scale personalized learning to diverse educational contexts.

Introduction

The rapid growth of Artificial Intelligence (AI), particularly Large Language Models (LLMs), has transformed educational technology by enabling systems that closely emulate human tutoring. Unlike traditional Intelligent Tutoring Systems (ITS) that depend on rigid rules and lack adaptability, LLMs can dynamically adjust explanations, assessments, and feedback based on real-time learner input. The paper presents a new AI tutoring framework integrating generative AI with learning analytics to deliver personalized, scalable, and data-driven instruction comparable to human tutoring.

Literature Review

Recent research highlights substantial progress in AI-based educational tools.

• Systems like Nexia Tutor use BERT models and gamification to support dyslexic learners.

• LLM-enhanced ITS have improved feedback quality, adaptability, and student engagement.

• Studies show generative AI can support programming education but caution against potential over-reliance.

• Other works emphasize ethical concerns such as bias, transparency, and data privacy.

• While many solutions offer personalization or domain-specific instruction, few achieve balanced personalization, scalability, and ethical robustness.
  The proposed framework aims to fill these gaps by combining generative AI, real-time analytics, and modular system design.

Methodology

The tutoring system is built using a modular microservice architecture with three main components:

1. User Interface – Developed with React and Next.js for interactive quizzes, explanations, and chat-based tutoring.

2. Backend Services – Built on FastAPI, handling user sessions and real-time communication without a dedicated database (in-memory processing).

3. AI Integration – LLMs such as Gemini 1.5 generate personalized explanations, adjust quiz difficulty using Bloom’s taxonomy and knowledge tracing, and provide emotionally adaptive feedback through sentiment analysis.

Results and Discussion

A four-week controlled experiment with 60 engineering students compared the system to traditional study methods.

• The AI group showed a 31% improvement in post-test scores, outperforming the control group’s 18% gain.

• User engagement was higher, with an average of 22 minutes per session.

• More than 80% of users expressed satisfaction, citing immediate feedback and intuitive interaction.

• System features such as quiz generation, chatbot guidance, and PDF/YouTube summarization supported personalized learning.

The findings confirm that integrating LLMs with learning analytics in a modular online environment enhances performance, engagement, and learner satisfaction. The adaptive and personalized nature of the system effectively addresses individual learning gaps while maintaining a scalable and accessible format.

Conclusion

This research presents a robust and scalable AI tutoring framework that synthesizes generative language models with real-time learning analytics. Through a carefully designed architecture and empirical validation, the system demonstrates notable improvements in student learning and engagement. Unlike traditional ITS, this model accommodates diverse learner needs and provides timely, relevant feedback. Moreover, continued exploration into ethical AI practices will be essential to ensure inclusive, secure, and transparent educational experiences.

References

[1] R. Forest, “Nexia Tutor?: An AI-Powered Language Personalized Learning System for Kids with Dyslexia and Reading Challenges,” 2024 Int. Mobile, Intelligent, Ubiquitous Comput. Conf., pp. 98–103, doi: 10.1109/MIUCC62295.2024.10783640. [2] X. Meng, “Enhancing Automated Feedback in Intelligent Tutoring Systems via Large Language Models and Data Distillation,” 2025 13th Int. Conf. Inf. Educ. Technol., pp. 275–279, 2025, doi: 10.1109/ICIET66371.2025.11046219. [3] H. Shahri, M. Emad, and N. Ibrahim, “Elevating Education through AI Tutor?: Utilizing GPT-4 for Personalized Learning,” 2024 15th Annu. Undergrad. Res. Conf. Appl. Comput., pp. 1–5, doi: 10.1109/URC62276.2024.10604578. [4] Z. Benedikt, M. Hobelsberger, and G. Socher, “Exploring the Role of Large Language Models as Artificial Tutors”, doi: 10.1109/EDUCON62633.2025.11016571. [5] L. Bonde, “A Generative Artificial Intelligence Based Tutor for Personalized Learning,” 2024 IEEE SmartBlock4Africa, pp. 1–10, doi: 10.1109/SmartBlock4Africa61928.2024.10779525. [6] R. Qureshi, S. Hajare, and P. Verma, “A Review on the Role of Artificial Intelligence in Personalized Learning,” 2024 Asia Pacific Conf. Innov. Technol., no. Ml, pp. 1–5, 2024, doi: 10.1109/APCIT62007.2024.10673706. [7] Q. E. A. Ratul, E. Zaman, S. Das, and M. Memari, “The Role of AI in Transforming Education?: A Systematic Review of Trends,” 2025 Intermt. Eng. Technol. Comput., pp. 1–6, 2025, doi: 10.1109/IETC64455.2025.11039329. [8] L. Frank, F. Herth, P. Stuwe, M. Klaiber, F. Gerschner, and A. Theissler, “Leveraging GenAI for an Intelligent Tutoring System for R?: A Quantitative Evaluation of Large Language Models,” 2024 IEEE Glob. Eng. Educ. Conf., pp. 1–9, doi: 10.1109/EDUCON60312.2024.10578933.

Copyright

Copyright © 2025 Mayuri Gawade, Samay Kumar, Sanskar Zine, Vedant Sanap, Samyak Lokhande, Sanika Luktuke. 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.