Abstract:

In the Italian school system, the acquisition of handwriting remains a fundamental requirement, creating a significant barrier for children with specific learning disorders (SLDs), particularly dyslexia. For these children, the gap between cognitive abilities and writing performance often leads to a cycle of frustration, anxiety and reduced self-efficacy. Professional interventions, although effective, are resource-intensive and often inaccessible due to years-long waiting times. Consequently, there is a need for accessible, at-home support that can provide appropriate feedback without the emotional and cognitive pressure of a classroom setting.

This work proposes Homework Tutor, a tool that promotes autonomy, motivation and literacy competence in learners with writing impairments. The primary aim is to leverage the interactions between Image Analysis and Large Language Models (LLMs) to create a stress-free and gamified environment for writing practice. Homework Tutor utilises Generative AI to provide context-aware and age-appropriate feedback, focusing on constructive explanations that not only motivate the child through a challenging activity, but also make orthographical knowledge accessible and easily digestible.

The tool can be accessed via any mobile device and operates through a child-friendly interface that enables children to take a picture of their homework. This visual input is then processed by a handwriting recognition module, which translates the content of the picture into text, thus allowing the integrated LLM to detect orthographic and syntactic deviations against the norms of the target language. The interface then presents the results in a list of mistakes that the child can navigate through. Each mistake has its own section, where the child can read both its correct form and a concise explanation of the writing norms involved.

To facilitate a deeper understanding, a conversational agent serves as an on-demand tutor, answering any follow-up questions and encouraging the user to rewrite the specific segment through unlimited attempts.

Homework Tutor is enhanced by gamified mechanics that reward sustained effort rather than accuracy, ensuring the child remains motivated throughout the session.

Ultimately, this work shows how AI-driven interfaces can be designed to foster resilience and confidence in children with SLDs, transforming the too often solitary struggle of doing homework into an engaging and supported journey. Furthermore, it highlights the versatile nature of Large Language Models, demonstrating their significant potential as adaptable tools for broader educational research and methodological development.

Keywords:

Specific Learning Disorders, children, Large Language Models, literacy development, generative AI, chatbot.