Didactical-Driven Teacher Assistant for Dimensional Modeling Course
A study published on arXiv (2607.22598) introduces a teacher assistant designed for a French university course focused on dimensional modelling, emphasizing a didactical approach. While the system employs large language models (LLMs), their function is limited to language processing, with pedagogical choices made by deterministic modules that manage intent detection, concept association, and didactic strategy selection. It functions independently of commercial LLM funding or GPU resources. An evaluation involving 195 real student inquiries reveals that conventional semantic retrieval fails to consistently retrieve necessary pedagogical content, highlighting the importance of this structured methodology.
Key facts
- arXiv paper 2607.22598
- French-language university course on dimensional modelling
- LLM acts solely as linguistic executor
- Deterministic modules handle intent detection, concept linking, didactic approach selection
- No commercial LLM budget or GPU infrastructure
- Evaluation on 195 authentic student questions
- Standard semantic retrieval insufficient for pedagogical content
Entities
Institutions
- arXiv