LLM-based Personal Travel Planner

Amandi Nimasha, Dushan Fernando, Oshadha Wijerathne

This research paper presents an innovative Personal Travel Planner that utilizes a sophisticated multi-agent architecture powered by Large Language Models (LLMs). At the core of our approach is the collection of comprehensive user data at the time of registration, which includes biographical details and personal preferences. This data is continuously updated to ensure our system adapts to changes in user needs and preferences. Our system employs advanced LLMs to craft travel itineraries that excel in relevance, feasibility, and personalization, meeting and often surpassing user expectations. By integrating a series of specialized agents, our planner adeptly manages user interactions from intent classification to detailed itinerary creation by harnessing real-time data from various external sources. The architecture is comprised of distinct agents responsible for user interface, context selection, planning, and constraint verification, ensuring each component seamlessly contributes to the overall functionality of the system. We evaluate our planner using rigorous methods that benchmark against leading models and employ diverse performance metrics focused on accuracy, relevance, and user satisfaction. Our findings demonstrate that combining LLMs with a multi-agent framework can transform the travel planning process, providing a user-focused, adaptable, and context-aware planning tool that represents a significant advancement over existing models