Agentic MailBot: Autonomous Multi-Agent Email Response System with Contextual Intelligence

In this paper, we propose a novel autonomous email response system that uses multi-agent architecture, LLMs, and retrieval-augmented generation (RAG) to generate appropriate answers to business emails without human input.It features an advanced classification system that sorts emails into two distinct types: Type A (Customer Service Questions) and Type B (Order Lifecycle Questions), allowing them to receive different processing routes.Type A emails make use of RAG-based retrieval from vector databases and Type B emails are based on ReAct agents with integration of SQL database toolkit.As shown in our implementation, we achieved 91.2% classification accuracy and 88.4% success rate in the whole pipeline.The framework is created using LangChain, LangGraph, Python, and Google Gemini 2.5 Pro.