AdaptLearn AI: A Generative AI Framework for Bidirectional Personalization in Education

Teacher burnout and student disengagement are twin crises crippling modern education, with 72% of educators overwhelmed by administrative tasks and 61% of students underserved by one-size-fits-all curricula. While generative AI tools like GPT-4 offer transformative potential, existing systems address only isolated needs—automating grading or personalizing content, but never both. This paper introduces AdaptLearn AI, the first framework to unify teacher support and student adaptation through bidirectional generative AI. Our architecture combines fine-tuned Large Large Language Models(LLMs) for dynamic content generation (e.g., adjusting STEM problems via few-shot learning) with Retrieval-Augmented Generation(RAG) pipelines for instructor task automation (e.g., Individual educational plan(IEP) drafting, rubric design). Pilot tests with 5 teachers and 15 students demonstrated a 47% reduction in lesson planning time and 22% improvement in quiz scores compared to non-AI workflows validating that harmonizing stakeholder needs amplifies efficacy. Three key contributions distinguish this work: (1) a novel dual-stakeholder architecture enabling real-time synchronization between learner progress and teacher tools, (2) open-source modules for GPT-4/Large Language Model Meta AI(LLaMA) integration, prioritizing reproducibility, and (3) design guidelines for ethical AI-Ed deployment, including FERPA-compliant anonymization and bias audits. Results challenge the AI-Ed community to move beyond siloed solutions, proving that generative AI can uplift classrooms holistically—not through replacement, but partnership. Code and datasets will be open-sourced to accelerate research toward equitable, human-centered educational tools.