AI Agents Leveraging RAG and MCP for Insurance Knowledge Management and Enterprise Workflow Automation

This thesis explores the use of Generative AI agents, Retrieval-Augmented Generation (RAG) systems, and the Model Context Protocol (MCP) to improve knowledge management within complex organizational contexts. Conducted at Reale Mutua Assicurazioni within the Data Science Center of Excellence, the work focuses on two Proofs of Concept: the POD Assistant, a multi-agent system designed to support project teams in managing documentation and interacting with corporate platforms such as Azure DevOps, and the CGA Copilot, a chatbot specialized in querying and interpreting lengthy and technical insurance documents (General Conditions of Insurance). The research combines multi-agent architectures orchestrated through LangGraph, hybrid retrieval pipelines with semantic re-ranking on Azure AI Search, and the integration of MCP for scalable tool interoperability. Evaluation results highlight the effectiveness of workflow orchestration and advanced retrieval techniques in delivering precise, contextualized, and traceable answers, outperforming traditional RAG approaches. The findings demonstrate how advanced AI techniques can be adapted to real corporate needs, balancing technological innovation and operational efficiency, while opening perspectives for future applications of agentic AI in the insurance sector.