PEAT: A Scalable LLM-Powered Tutoring System with Real-Time Adaptation and Explainable Feedback for Diverse Learners
The Personalized tutoring systems have always tried to closely mimic tutoring by human beings in computerbased environments, but most fall short due to low levels of adaptability, opaque feedback, and difficulty scaling. In response to these issues, PEAT proposes a tutoring system based on LLMs that is scalable, provides real-time adaptive feedback and instruction that is personalized and explainable to various learners. It is a hybrid between large language model generation and reinforcement learning to dynamically adjust the prompts, examine learner profiles, and provide multi-layered feedback. On the large-scale datasets, PEAT scored up to 96.2 % and feedback explainability was higher than 0.95, and increased learner engagement by 61.4 % among different levels of education. It was also shown that the system re-adjusted quickly within less than 130 ms and exhibited great accuracy to reach correction rates of multi-step problems. Development Using EduSim++, PEAT was found to be much more adaptable, transparent, and responsive than the earlier systems. These findings make PEAT a next-generation technology that can democratize intelligent tutoring to become inclusive, real-time time and effective education delivery.
Paper
Full text
PEAT: A Scalable LLM-Powered Tutoring System with Real-Time Adaptation and Explainable Feedback for Diverse Learners
Semantic Scholar · 2025
Abstract
The Personalized tutoring systems have always tried to closely mimic tutoring by human beings in computerbased environments, but most fall short due to low levels of adaptability, opaque feedback, and difficulty scaling. In response to these issues, PEAT proposes a tutoring system based on LLMs that is scalable, provides real-time adaptive feedback and instruction that is personalized and explainable to various learners. It is a hybrid between large language model generation and reinforcement learning to dynamically adjust the prompts, examine learner profiles, and provide multi-layered feedback. On the large-scale datasets, PEAT scored up to 96.2 % and feedback explainability was higher than 0.95, and increased learner engagement by 61.4 % among different levels of education. It was also shown that the system re-adjusted quickly within less than 130 ms and exhibited great accuracy to reach correction rates of multi-step problems. Development Using EduSim++, PEAT was found to be much more adaptable, transparent, and responsive than the earlier systems. These findings make PEAT a next-generation technology that can democratize intelligent tutoring to become inclusive, real-time time and effective education delivery.