AI-Driven Learning Assistants for Reflective, Adaptive, and Inclusive Education

This thesis presents the AI-Driven Reflective Learning Framework (AI-DRLF), a framework for the design of educational assistants based on Large Language Models (LLMs). The framework is intended to support reflective, adaptive, and inclusive learning through the integration of generative AI technologies and pedagogical principles. In particular, it combines reflective scaffolding, adaptive content delivery, accessibility-aware interaction, and AI-supported feedback within a unified educational architecture. To explore and evaluate the framework, several experimental systems were developed, including conversational learning assistants, concept-mapping environments, accessibility-oriented tools for students with dyslexia, and AI-supported evaluation systems. The proposed approach integrates Retrieval-Augmented Generation (RAG), fine-tuning, multimodal interaction, and readability-aware language generation with pedagogical models such as Cognitive Load Theory, the ICAP framework, and the Improved Vee Heuristic. The experimental results indicate that the AI-Driven Reflective Learning Framework can support learner engagement, accessibility, personalization, and metacognitive reflection in AI-based educational environments. The thesis therefore provides both a conceptual framework and a set of practical implementations for the development of human-centered AI systems intended to support teaching and learning processes.