Date of Publication : May 2026

Abstract: Museums serve as critical repositories of human cultural heritage, yet the traditional mechanisms for visitor engagement— relying heavily on static placards, manual audio guides, and simplistic QR codes—often fail to deliver personalized, interactive, or sufficiently deep educational experiences. Visitors are frequently left unable to ask follow-up questions or effortlessly identify artifacts without locating physical identifiers. This paper introduces the Intelligent Augmented Reality (AR) Museum Guide, a novel, web-native framework that synthesizes two advanced artificial intelligence paradigms to redefine cultural heritage interaction. First, we deploy a fine- grained visual recognition engine, engineered via transfer learning on the ConvNeXt-Base architecture, enabling the instantaneous identification of artifacts from user-captured smartphone images without requiring physical markers. Second, we integrate a domain constrained Retrieval-Augmented Generation (RAG) pipeline, leveraging ChromaDB and Google Gemini Flash Lite, to provide dynamic, conversational question-answering capabilities strictly grounded in curator-verified documentary sources. Evaluating our system on a comprehensive custom dataset of 20 artifact classes (480 images), the vision model achieves a top-1 validation accuracy of 94.4% and a top-3 accuracy of 99.1%, significantly outperforming baseline convolutional and transformer models. Concurrently, the RAG subsystem, evaluated across 80 complex historical queries, demonstrates a 0% hallucination rate while maintaining 100% source-level citation accuracy. By eliminating the friction of mobile application installation and guaranteeing cryptographic-level factual grounding, this architecture presents a scalable, highly interactive blueprint for the next generation of intelligent museum spaces.

Reference :