International Journal of Scientific Research and Engineering Development

Large Language Models (LLMs) show strong capabilities in language understanding and reasoning, but face key limitations in education, including hallucinations, lack of domain specificity, outdated knowledge, and privacy concerns. These issues affect reliability when applied to institutional academic data. This paper proposes RAG-AI, a multi-modal Retrieval-Augmented Generation framework for context-aware and privacy-preserving academic assistance. The system combines Whisper-based speech recognition, semantic chunking, BGE-M3 embeddings, cosine similarity retrieval, and locally deployed LLMs. By retrieving relevant institutional data before response generation, RAG-AI reduces hallucinations and improves accuracy. Experiments on lecture recordings and academic PDFs show improved factual reliability compared to standalone LLMs. The system supports local deployment, ensuring data privacy and compliance, and provides a scalable solution for educational AI.

Pratik S. Bhiwarkar, Khushi S. Telgote, Lokesh J. Pande, Shantanu D. Gawande, Falgun A. Kawale, Prof. Mukund Joshi,"RAG-AI: A Multi-Modal Retrieval-Augmented Generation Framework for Academic Institutions" International Journal of Scientific Research and Engineering Development, V9(2): Page(1378-1383) Mar-Apr 2026. ISSN: 2581-7175. www.ijsred.com. Published by Scientific and Academic Research Publishing.