Large Language Models (LLMs) for Generating Personalized Educational Feedback: Architecture, Methods, and Perspectives
TL;DR
This work proposes SIGFE (Smart Intelligent Generation of Feedback for Education), a modular framework that produces such feedback with LLMs that scales and deals with bias, transparency, and learner agency and sets out what remains to be done on real-time and multimodal delivery.
Abstract
Recent advances in large language models (LLMs), such as GPT-4 and LLaMA, have opened new opportunities for personalized learning support in intelligent educational environments. Feedback matters for how students stay motivated, how they study, and how well they do, but giving each of them personalized feedback in real time is rarely workable: someone, or something, has to answer thousands of students at once. We propose SIGFE (Smart Intelligent Generation of Feedback for Education), a modular framework that produces such feedback with LLMs. It has six components: data collection and preprocessing, learner profile analysis, content-to-goals mapping, LLM-based feedback generation, personalization and adaptation, and an evaluation loop that combines human assessment with usage logs. We tested it in domain-independent LLM-based settings on six real learner-record datasets taken from a Learning Management System. On both quantitative and qualitative measures, the feedback produced by SIGFE was more accurate, more informative, better contextualized, and more personalized than the output of a rule-based system. Learner modeling and prompt engineering turned out to matter most for how well the LLM performs in this setting. For AI in Education, the work offers a personalized feedback mechanism that scales and deals with bias, transparency, and learner agency, and it sets out what remains to be done on real-time and multimodal delivery. Received: 7 June 2026 | Revised: 17 August 2026 | Accepted: 23 September 2026 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement The data that support the findings of this study are openly available in GitHub at https://github.com/riiid/ednet. Author Contribution Statement Ridouane Oubagine: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization. Ibtissam Azzi: Conceptualization, Software, Validation, Writing – review & editing, Supervision. Loubna Laaouina: Methodology, Formal analysis, Writing – original draft, Supervision, Project administration. Adil Jeghal: Conceptualization, Methodology, Validation, Investigation, Resources, Data curation, Writing – review & editing, Visualization, Supervision, Project administration. Hamid Tairi: Software, Visualization, Supervision, Project administration.