Retrieval-Augmented Generation for Knowledge-Driven Question Answering in Traditional Chinese Medicinal Diets

Large language models (LLMs) have demonstrated great potential in open-domain question answering. However, their generated content is often affected by hallucinations and outdated knowledge, limiting their reliability in specialized domains. To improve the quality of question answering in the field of traditional Chinese medicinal diets, this study designs and implements a Retrieval-Augmented Generation (RAG)-based intelligent QA system. Built upon a corpus of medicinal diet literature, the system employs tool-augmented LLMs to automatically extract knowledge triples, and constructs an efficient retrieval mechanism through semantic clustering and Bloom filters. On top of this, a personalized response agent is introduced to dynamically integrate retrieval results based on the user’s body constitution and query intent, enabling the model to generate responses more tailored to individual needs. Experiments evaluate the performance differences of several mainstream models in both knowledge extraction and answer generation, providing practical insights and technical references for building trustworthy, scalable, and personalized QA services in the medicinal diet domain.