This study introduces an advanced intelligent reader aimed at improving online extracurricular reading for visually impaired learners, surpassing existing tools by offering intelligent information processing. Utilizing large language models (LLMs), our system delivers personalized content recommendations and summaries based on users’ historical interests, a departure from conventional approaches. It features a novel prompt optimizer that enhances the system’s ability to tailor information to individual preferences through an iterative bootstrapping process. This paper highlights the system’s dual innovations in personalized content summarization and LLM-based adaptive information recommendation, demonstrating its potential to make online learning more accessible and engaging for visually impaired students. In addition, it explores the challenges and ethical considerations of integrating LLMs into assistive technologies, contributing valuable insights for future developments in adaptive content delivery.