DIANGPT: A Review on a Domain-Specific Question Answering System Using LORA and RAG

DianGPT is a domain-specific question-answering (QA) framework designed to transform institutional archival content into an accurate, context-aware conversational knowledge system. Integrating Low-Rank Adaptation (LoRA), Retrieval- Augmented Generation (RAG), FAISS-based retrieval, and struc- tured dataset engineering, the system optimizes both precision and computational efficiency. This review evaluates DianGPT’s architectural pipeline, including dataset processing, supervised fine-tuning, retrieval workflows, and automated evaluation strate- gies. Further, the paper critically analyzes the reported ex- perimental outcomes, highlights methodological strengths, and identifies gaps and potential improvements. The work serves as a comprehensive reference for researchers developing lightweight, domain-optimized large language models.