AI-Based Intelligent Tour Guiding and Tourism Recommendation System Specialized for Pune

Abstract

Tourists visiting Pune rely on multiple disconnected sources — directories, map apps, review platforms — to plan a visit, making discovery slow and unreliable. This paper presents the design, implementation and evaluation framework of an integrated AI tourism platform for Pune combining a source-verified geospatial tourism database (PostgreSQL/PostGIS), a deterministic-first personalized recommendation engine, a deterministic itinerary-planning heuristic, and a Retrieval-Augmented Generation (RAG) conversational layer (Gemini + pgvector) that grounds AI answers in verified data. The system (Spring Boot/Angular) currently holds 18 verified places, 21 categories and an 8-chunk starter knowledge base; its geospatial and vector-search mechanics were live-verified, including a genuine indexing bug found and fixed. A synthetic Stage 1 dataset of 120 responses covering seventeen of the system's seeded Pune-area places was analyzed (synthetic data was used for Stage 1 owing to time constraints and will be replaced by real reviewer data in Stage 2); the illustrative results show high overall visit satisfaction (mean 3.87/5) and a strong preference for personalized recommendation (mean importance 4.30/5), alongside crowd comfort as the lowest-rated experience factor (mean 3.43/5), consistent with overcrowding being the most frequently reported visitor complaint. The paper's contribution is an architectural and methodological integration of verified tourism data, geospatial computing and grounded conversational AI for a single Indian city, evaluated with Stage 1 synthetic data and reported with an explicit account of what is implemented-and-verified versus proposed.