Abstract: Currently, large language models have become the focus of society. However, due to the diversity of engineering projects, the construction industry's demand for knowledge retrieval shows significant differences. Based on the above phenomenon, this study leveraged Retrieval-Augmented Generation (RAG) technology to propose a design framework for constructing an intelligent question-and-answer system for the construction field. Through technologies such as metadata tagging, document segmentation and aggregation, and vectorized storage, a dedicated knowledge base and a keyword repository were constructed. By integrating business logic rules, it achieved personalized customization of output content, thereby addressing the heterogeneous requirements of multiple project stakeholders and subsidiary units under group corporations. Besides, the system relied on multi-dimensional security protection system, including sensitive data filtering, safety classification and isolated network retrieval, to effectively ensure information security in the knowledge retrieval process. The research outcomes provide technical support for reducing information retrieval costs and advancing the construction industry's transition toward data-driven knowledge service models.