AI-powered chatbots are increasingly used in education to provide personalized support, especially in large-scale and online learning environments. However, concerns remain regarding the factual accuracy of generative AI responses and potential over-reliance by students. This study implemented a Retrieval-Augmented Generation (RAG)-based chatbot in a Digital Signal Processing course, leveraging both lecture materials and student-contributed study articles. A comparative experiment with four configurations, using lecture knowledge, student knowledge, both, or neither, revealed that the hybrid approach consistently produced the most accurate and contextually relevant responses. Our findings demonstrate that combining structured academic content with student-contributed contextual insights enhances chatbot performance. This study offers empirical guidance for designing AI-enhanced educational tools that promote engagement and deeper understanding.