Optimized Travel Recommendation & Itinerary Generation

Abstract: The rapid growth of digital travel platforms has created an urgent need for intelligent, personalized travel recommendation systems that go beyond simple keyword-based search. This paper presents TripSona, a full-stack AI- powered Indian travel recommendation and itinerary planning system built on a microservices architecture. The system employs a hybrid recommendation engine that combines a trained Machine Learning model using collaborative and content-based filtering with real-time data from Google Places API and Booking.com API to generate personalized destination recommendations. For itinerary generation, the system introduces a novel Multi-Agent Architecture comprising six specialized Gemini AI agents — Day Planner, Weather, Budget, Food and Tips, Transport, and Coordinator — each responsible for a distinct aspect of trip planning. The system is trained and evaluated on a dataset of 2000 Indian travel records spanning 40 destinations across diverse categories including Beach, Adventure, Nature, Historical, and City. User preferences including age range, budget, trip duration, interests, purpose, health conditions, and cuisine preference are used as input features. Experimental results demonstrate that the hybrid recommendation approach achieves superior personalization accuracy compared to standalone ML or API-based approaches. The multi- agent itinerary system produces contextually rich, weather-aware, budget-conscious day-by-day travel plans with real venue data. The system is deployed as a React-based web application integrated with Firebase Authentication and PostgreSQL for persistent storage.

Keywords: Travel Recommendation System, Multi-Agent Architecture, Machine Learning, LangChain, Gemini AI, Google Places API, Microservices, Collaborative Filtering, Content-Based Filtering, Indian Tourism