Reclaiming Pedagogical Control in DSP Education through Hybrid-RAG-Enhanced Local LLM

The growing use of LLMs by students as a substitute for textbooks introduces important pedagogical challenges. Many LLM-generated answers do not follow the notation, methodology, or conceptual approach adopted by instructors, and may even contain errors. This shifts educational influence from professors to Big Tech platforms that define model behavior and collect interaction data. Furthermore, they impose an interaction barrier that relies on online services, which can lead to latency in human-machine communication. This paper presents both a quantitative and qualitative evaluation of this problem and proposes an open-source AI chatbot tutor based on locally hosted LLMs and a Hybrid Retrieval-Augmented Generation (Hybrid RAG). The system uses course-specific material about digital signal processing created by the authors, preserving privacy and enabling cost-controlled deployment within institutional infrastructure.