This article introduces JudoMath, a semantic-based intelligent tutoring system designed to support the development of arithmetic skills among French elementary school students (Cycle 3). The system relies on an ontological modeling of mathematical competencies aligned with the Eduscol national curriculum, enabling the generation of personalized exercises and real-time tracking of student progress. JudoMath’s architecture combines a symbolic AI engine (OWL ontologies, SPARQL queries, inference rules) with gamified elements (badges, belts, personalized avatars), aiming to provide a transparent, interpretable, and engaging learning experience.

An exploratory field study was conducted in an extracurricular setting involving 32 students from four schools. The experiment focused on evaluating the usability, user engagement, and perceived quality of the system, rather than measuring learning outcomes. Mixed-method data were collected through system logs and the AttrakDiff questionnaire, adapted for young users. Results showed a generally positive reception: students used the system autonomously, reported enjoyment and clarity of the interface, and appreciated the belt-based progression system. Average usage time per student was 23.8 minutes per session, with repeated use across several weeks.

Qualitative feedback from students and educators highlighted several strengths, including intuitive navigation, motivational design, and the potential for autonomous learning beyond classroom content. However, limitations emerged regarding the variety of content, the depth of feedback, and the absence of social features. Educators expressed a need for more detailed explanations, audio support for non-readers, and the ability to create custom exercises.

Compared to other tutoring systems (e.g., ALEKS, Khan Academy, Duolingo Math), JudoMath stands out through its explicit pedagogical alignment, high transparency, and symbolic reasoning capabilities. While connectionist AI systems often offer powerful performance, they lack explainability and adaptability from a teacher’s perspective. JudoMath addresses this gap by allowing educators to inspect and modify the rules governing student progression.

This proof-of-concept study demonstrates the technical feasibility and educational relevance of a semantic tutoring system in real-life settings. Future developments will focus on expanding content diversity, improving feedback mechanisms, and introducing collaborative activities. A controlled study with pre- and post-tests is planned to assess the actual impact on learning outcomes. Additionally, hybrid approaches combining symbolic AI and machine learning (e.g., error clustering, NLP for open responses) are considered to enhance adaptability while preserving explainability.

Keywords: Intelligent Tutoring System (ITS), Semantic Technologies, Ontologies, Arithmetic Learning, Personalized Learning, Symbolic AI, Eduscol Curriculum, User Experience, Educational Technology, Explainability, Learning Analytics.