基于本地化大模型与RAG技术的银行业智能客服系统研究与应用

Research and Application of an Intelligent Customer Service System for the Banking Industry Based on Localized Large Models and RAG Technology

摘要: 为解决传统金融客服模式响应效率低、服务个性化不足及普惠客群覆盖有限等问题，特别是助力中小银行在普惠金融服务中突破技术投入有限、专业人才短缺等数字化转型困境，本研究设计并实现了一套全链路本地化部署的智能客服系统(AI Agent)。该系统采用Ollama部署大语言模型(LLM)作为核心推理引擎与向量嵌入模型，结合Xinference部署自动语音识别(ASR)、文本转语音(TTS)与重排序(Rerank)模型，并基于Dify平台实现可视化工作流编排、内容审查与应用程序接口(API)服务管理。同时，集成RAGFlow构建本地私有知识库，对金融文档进行向量化处理与检索，形成基于检索增强生成(RAG)的技术框架。结果表明，该系统在保障金融级安全合规的前提下，显著提升了客服响应效率，实现了7 × 24小时全天候运行，并通过智能化客户洞察与个性化服务，有效增强了中小银行在普惠金融领域的服务能力与竞争力。

Abstract: To address the issues of low response efficiency, insufficient personalized services, and limited coverage of inclusive customer groups in traditional financial customer service models, especially to help small and medium-sized banks overcome the digital transformation challenges such as limited technical investment and shortage of professional talents in inclusive financial services, this study designs and implements a fully localized and end-to-end deployed intelligent customer service system (AI Agent). The system adopts Ollama to deploy large language models (LLM) as the core inference engine and vector embedding model, combines Xinference to deploy automatic speech recognition (ASR), text-to-speech (TTS), and re-ranking (Rerank) models, and is based on the Dify platform to achieve visual workflow orchestration, content review, and API service management. At the same time, it integrates RAGFlow to build a local private knowledge base, vectorizes and retrieves financial documents, forming a technology framework based on retrieval-augmented generation (RAG). The results show that the system significantly improves the response efficiency of customer service while ensuring financial-level security and compliance, operates 24/7, and effectively enhances the service capabilities and competitiveness of small and medium-sized banks in the inclusive finance field through intelligent customer insights and personalized services.