RAG-Enhanced AI Agent Chatbot Architecture for Optimizing Insurance Operations

In life insurance, customer inquiries often involve complex terminology and evolving policy content. Traditional rule-based systems struggle to meet these demands, and manual services are costly and difficult to scale. To address this, this paper proposes a RAG-enhanced AI agent chatbot that integrates the Ollama language model with the Supabase vector database, achieving a deep fusion of language generation and knowledge retrieval for complex insurance scenarios. Constructed POC test set of 50 real-world questions, covering four core subcategories: Insurance Laws and Regulations, Policy Administration and Contract Validity, Interpretation of Insurance Clauses, and Principles and Fundamentals of Insurance. Experimental results show that the application outperforms traditional AI applications in semantic understanding, term adaptation, and logical consistency. It better satisfies industry needs for answer traceability and contextual integrity, demonstrating strong practical value. Overall, from the perspective of cost-effectiveness, the evaluation demonstrates that a “lightweight model with highquality RAG retrieval” architecture excels in knowledgeintensive tasks, with performance more dependent on knowledge path quality than model size. This architecture highlights the synergy between the Lightweight model and RAG, offering a feasible path and key reference for intelligent insurance customer service and business automation.