HyPAM-ITS: a Hybrid Pedagogically-Align Intelligent Tutoring System Multi-Verifier

Abstract

In the modern educational environment, the issue of delivering quality and individualized education to the ever increasing student population around the world has become an issue of critical proportions. The old-fashioned classroom model will not be able to scale, but the recent breakout of Artificial Intelligence (AI) and Machine Learning (ML) presents an opportunity. Nevertheless, the implementation of AI in education is also frequently negated by the problems of accuracy, hallucination, and the lack of pedagogical disturbances. We introduce HyPAM-ITS, a new, AI-based intelligent tutoring platform that reinvents the online learning process. HyPAM-ITS uses a complex, multi-level Large Language Model (LLM) pipeline, that is, a combination of Gemini 1.5, Gemini 1.5 Pro and Gemini Pro in a scaled MERN-stack architecture, in contrast to a typical chatbot, or a rule-based, fixed system. The system is not only created to answer questions, but is expected to teach, and thus, using Retrieval-Augmented Generation (RAG) to base its answers on validated course materials, factual integrity is maintained. Moreover, it has an effective feedback loop that is real-time and enables the system to change the teaching approach depending on the level of understanding and emotion of the student. The result of our comprehensive tests that have tested hundreds of complicated academic queries in the areas of Science, Technology, Engineering, and Mathematics (STEMs) shows that this multi-model fallback model has a total accuracy of about 95%. Importantly, the system minimizes the occurrence of harmful AI hallucinations to only 3.2 percent, which is a credible, emotionally conscious, and safe environment to the contemporary learner. This study confirms this hypothesis: a human-in-the-loop architecture with tiered AI verification is the secret to sustainable and equitable education technology.