Abstract

Large Language Models (LLMs) have shown strong abilities in understanding and generating natural language, but they often
produce inaccurate or misleading responses when applied to specialized educational domains. This paper introduces an
Educational Retrieval-Augmented Generation (RAG) system tailored for textbook-based learning, where accuracy is critical.
Instead of depending completely on the model's internal knowledge, the method recovers appropriate knowledge from a
curated collection of domain-specific textbooks. It utilizes a fusion recovery methodology that links keyword matching with
semantic similarity to detect the most appropriate content, which acts as context for a locally deployed LLM to make precise
answers. Developed as a web-based platform using frameworks like Django, the system minimizes dependency on external
APIs and includes a preprocessing mechanism to convert visual elements like diagrams into descriptive text. Overall, the
system provides accurate, explainable, and domain-focused educational support.