CONFERENCE / ICCAIS-2026

Intelligent Tutoring Systems (ITS) have received a significant amount of focus due to the development of artificial intelligence in an effort to facilitate learning on a more personalized and interactive level. However, the majority of existing AI-based tutoring systems run on a single large language model and each of them is usually cursory to hallucinating responses, curriculum disconnection, and lack of transparency in reasoning. This paper describes an intelligent tutoring system, which uses AI, with multi-LLM orchestration, including retrieval- augmented generation and knowledge graphs in the context of delivering the correct, definable, and curriculum-aware academic assistance. The proposed system chooses the queries of the learners selectively for special language models and grounds the responses with the assistance of the trusted external resources based on a retrieval-augmented pipeline and uses a curriculum- based knowledge graph to identify the gaps in skills and generate personal learning paths. Additionally, agentic reasoning systems built around ReAct and optimized Tree-of-Thoughts are useful in providing step-by-step reasoning and easy decision-making, and course-specific fine-tuned small language models are useful in increasing domain accuracy and efficiency. The outcomes of the experimental observations made with the assistance of the system- level analysis to determine the effectiveness of personalization, the relevancy of responses and the increased level of engagement with a learner demonstrate the fact that the proposed framework can potentially achieve scalable and reliable intelligent tutoring.