Abstract

Large Language Models (LLMs) have brought about a fundamental shift in how humans interact with knowledge systems. Despite their impressive fluency, standard LLMs are prone to generating plausible sounding but factually incorrect responses a phenomenon widely known as hallucination especially in specialized technical domains such as engineering and computer science education. This paper proposes a Retrieval-Augmented Generation (RAG) based Question Answering (QA) system designed specifically for technical education. The system ingests engineering textbooks, syllabus PDFs, and research papers, converts them into dense vector embeddings stored in a FAISS index, and grounds each LLM response in semantically retrieved context. Experimental evaluation against a baseline LLM reveals substantial improvements: Precision increased by 52%, Recall by 16%, BLEU score by 27%, ROUGE-1 by 42%, and F1-score by 45%. The results confirm that retrieval-augmented generation is a practical and effective strategy for building reliable, domain-specific educational QA systems with meaningfully reduced hallucinations.