AI-Based Tourism Recommendation and Itinerary Planner: A Retrieval Augmented Approach to Personalized

Planning a multi-city trip within an Indian state remains a fragmented exercise. Attraction details, opening hours, indicative visit durations and accessibility information are scattered across official portals, travel blogs and word-of-mouth advice, and generic itineraries rarely respect a traveller’s time, budget, mobility or language preferences. This paper presents the design of an AI-Based Tourism Recommendation and Itinerary Planner, a Retrieval-Augmented Generation (RAG) prototype that converts curated regional tourism records into personalized, day-wise itineraries, each recommendation carrying an explanation and a visible source. The system is organised as five independently verifiable layers—data, retrieval, intelligence, application and responsible AI—realised with PostgreSQL, PGVector, Spring Boot, Spring AI and a lightweight web interface. A hybrid retrieval stage combines semantic similarity with metadata filters; a transparent weighted scoring function ranks candidate places; and a time-budgeted composer distributes them across days while limiting inter-city movement. A grounding verifier withholds any statement that cannot be traced to the knowledge base, and an Indic-language layer built around Sarvam-style translation and speech services extends access to Hindi and Hinglish speakers. The prototype deliberately excludes live booking and availability checks, and any data that cannot be verified from a public source is labelled as mock. The paper describes the framework, its algorithms and an evaluation protocol, and argues that layered, source-grounded design is a practical route to trustworthy AI assistance for regional tourism information centres.