FURI | Summer 2025

Domain-Specialized Course Assistant: Fine-Tuning and RAG-Enhanced LLMs for Precision in Cloud Computing Education

Generic large language models (LLMs) often provide inaccurate or irrelevant answers to course-specific questions, limiting their educational value. To solve this, we developed a specialized AI assistant for ASU’s Cloud Computing course by fine-tuning a local DeepSeek-R1 model (deployed via Ollama/Docker) on professor-student Q&As and integrated RAGFlow for retrieval-augmented generation from lectures, syllabi, and validated academic sources. This approach yielded higher accuracy and reduced hallucinations compared to generic LLMs for course content.