2026 International Conference on Networks Computers and Communications (ICNCCom 2026)

Preparation for mathematical Olympiads like Regional Mathematical Olympiad (RMO) would be hard without any form of structured training, personal tutoring, and learning resources. In this paper, we have designed an intelligent tutoring system using AI tutor for solving complex mathematics questions with retrieval augmented generation (RAG), multimodal inputs, symbolic validation, and personalized learning methods. The system makes use of math OCR to process handwritten inputs, coordinate requests through the FastAPI webserver, perform vectorized semantic search with PostgreSQL database with pgvector support, and finally employ large language models to hint, evaluate, and tutor students. Instead of providing direct answers, the tutor uses the Socratic method to guide students. Some features of the system include problem generation, personalized recommendations, step-by-step grading of solutions, and monitoring of the learning process. Accuracy achieved by the system is 86% in final answers, 81% in logical reasoning, 89% in grading, and 91% in math OCR on a selected sample of 120 questions from the Omni-MATH database. Positive feedback was received from the preliminary trials conducted for RMO aspirants.

Mathematical Olympiad,AI Tutor,Retrieval-Augmented Generation,Symbolic Verification,Math OCR,Personalized Learning,Olympiad Preparation,Educational Technology