Abstract: In response to the frequent training and repeated questioning of users during the promotion and application of the Financial Shared Service Center and financial shared service management information system, this paper designed a financial local think-tank system (referred to as the local think-tank system) based on a large language model, built the overall architecture of a local think-tank system by deploying a domestic open-source large language model locally on the intranet, combined with the LangChain framework, and elaborated on the system's functions and key technologies in detail. The system was relied on the document resources uploaded from the vector database and fine tuned and optimized in the later stage, could provide a generative dialogue and question answering service to assist the staff of the Financial Shared Service Center in conducting online and offline training work. Experimental verification shows that the system can accurately answer the questions raised by users, reduce the workload of training and promotion personnel, and also has application prospects in other railway professional training work.