This paper focuses on providing effective and accessible learning experiences by interpreting various question types in AI-based education. To overcome the limitations of existing systems, this paper presents Uni-Retrieval, a lightweight multi-modal retrieval module. Uni-Retrieval extracts question type prototypes and dynamically matches them with tokens in a continuously updated Prompt Bank. The Prompt Bank, which encrypts and stores domain-specific knowledge using the MoE-LoRA module, is used to enhance the ability of Uni-Retrieval to adapt to unseen question types at test time. By integrating Uni-Retrieval with an instruction-tuned language model, we build a complete retrieval-augmented generation pipeline called Uni-RAG, which retrieves relevant training materials and generates explanations, feedback, or training content that aligns with learning objectives given style-conditional questions. Experimental results on SER and other multi-modal benchmarks demonstrate that Uni-RAG outperforms baseline retrieval and RAG systems in both retrieval accuracy and generation quality while maintaining low computational cost. This study provides a scalable, pedagogically informed, intelligent teaching system that links discovery and creation to provide personalized, explainable, and efficient learning support in a variety of STEM scenarios.
