BUILDING AN INTELLIGENTSTUDYASSISTANT APPLICATION

Abstract

This thesis addresses the growing need for personalized, efficient, andreliablelearning support in modern education. The research problem focuses on improvingtheaccuracy and contextual relevance of automated academic assistance. Totacklethischallenge, the study proposes a smart learning system that integrates a largelanguagemodel with a Retrieval-Augmented Generation (RAG) framework. The methodologyincludes constructing a vector database using FAISS, applying semantic-awaredocument chunking, and employing advanced embedding models for retrieval.Experimental evaluations show that the system delivers more relevant, faithful, andcontext-grounded responses than baseline keyword-based and rule-basedapproaches.The results demonstrate that RAG-enhanced LLMs significantly reduce hallucinationswhile improving instructional precision. The thesis concludes that the proposedarchitecture is scalable, adaptable to various academic domains, and offers apracticalpathway for deploying AI-powered tools in educational environments.