AI-Driven Hyperpersonalization in Tourism: A Conversational Agent Using Reinforcement Learning for Dynamic Itinerary Recommendations

The tourism industry increasingly requires adaptive recommendation systems capable of understanding traveler preferences and learning from user behavior. Traditional conversational agents and retrieval-based systems remain limited by static knowledge and lack continuous personalization. This study presents an integrated framework combining Large Language Models, Retrieval-Augmented Generation, and Reinforcement Learning via Multi-Armed Bandit algorithms to develop an adaptive conversational assistant for dynamic travel itinerary recommendations. Orchestrated with LangGraph for stateful dialogue management and deployed on AWS, the system continuously updates its recommendation policies from user interactions. Case studies demonstrate its ability to balance exploration and exploitation. The conversational design emulates human travel advisors through guided questioning and preference elicitation. This architecture bridges the gap between static retrieval and adaptive learning, enabling a closed-loop conversational AI that evolves from user feedback. The research validates the feasibility of production-ready, reinforcement learning–driven personalization in tourism and provides a scalable foundation for adaptive recommendation systems beyond tourism.