Development of an Automated System for Generating Assignments in Mathematical Disciplines Using LLMs
V. I. Belousova, Denis Gorbunov, Михаил Алексеевич Ерин · 2025
The paper discusses the implementation of a three-level model ("Advanced-Basic-Elementary") for teaching mathematics to students on individual trajectories at IRIT-RTF. It highlights the need for adapting courses to meet diverse student backgrounds, particularly focusing on students with lower levels of mathematical preparation or those facing learning challenges. The authors introduce the elective course "Elementary Fundamentals of Mathematics," designed to address gaps in foundational knowledge and improve academic success rates. A key innovation presented is the development of an automated system for generating unique test tasks using modern AI technologies, including retrieval-augmented generation approaches and large language models. The system ensures both the educational relevance and uniqueness of generated tasks, addressing issues related to academic dishonesty and inefficiencies in traditional testing methods. Its integration with Wolfram Alpha API enhances the accuracy of answer calculation. The study outlines the methodology behind the system’s development, emphasizing its adaptability across various mathematical topics. Results demonstrate effectiveness in improving the quality of knowledge assessments and reducing reliance on proprietary tools like Mathpix and Wolfram Alpha. Future plans include refining the system through model fine-tuning and expanding its capabilities. The research contributes significantly to the automation of knowledge control processes, enhancing flexibility and precision in assessing students’ progress.