Adaptive Reading Material Generation System Based on Large Language Models: A New Exploration in Educational AI

The master's thesis project from Tampere University in Finland built an intelligent system that uses Large Language Models (LLMs) to dynamically generate reading materials and adaptively adjust difficulty, aiming to address the personalized learning pain point of traditional textbooks' "one-size-fits-all" approach. The system supports OpenAI and local open-source models (via Ollama), with core functions of dynamic generation and adaptive difficulty control, providing an innovative application paradigm for the field of educational AI.

In language learning and reading education, traditional textbooks struggle to meet personalized needs: advanced learners find the content too simple, while beginners are easily discouraged by overly difficult materials. This project directly addresses this educational challenge and proposes a solution using LLMs to dynamically generate and adjust reading materials.

The system adopts a front-end and back-end separation architecture: the front-end is based on the Next.js framework, and the back-end uses Express.js to build a RESTful API. Technically, it supports a dual-mode option of OpenAI GPT series or Ollama local models. Core functions include generating customized materials by receiving parameters such as topic, difficulty, word count, and language; the adaptive mechanism identifies language features through prompt engineering and dynamically adjusts the difficulty of subsequent materials based on learners' performance.

Taking the topic of "rainforest ecosystem" as an example, the system can generate content of different difficulty levels: texts for beginner learners focus on basic vocabulary and simple sentence structures; texts for advanced learners include professional terms and complex arguments, breaking through the limitations of traditional static textbooks.

The system is applicable to multiple scenarios: teachers can quickly generate layered teaching materials; self-learners can avoid comfort zones or frustration; those with special educational needs (such as dyslexia, gifted students) can obtain customized content, demonstrating unique value.

The system has strong scalability and can integrate models such as Anthropic Claude and Google Gemini; in the future, functions like audio reading, vocabulary annotations, and reading comprehension question generation can be added. This project represents the trend of educational AI shifting from static content to dynamic generation, promoting the transformation of personalized learning.