Triangolazioni's LLM Guide Offers Context-Aware Cultural Heritage Info

Researchers at the University of Turin have extended Triangolazioni, a web-based cultural heritage guide, with large language models to deliver context-aware information to visitors. The system, described in a paper accepted at the ACM UMAP '26 conference, aims to enrich curated cultural content with dynamic, external knowledge that adapts to the user's current context—such as location, interests, or specific queries. The architecture is designed to be loosely coupled and agnostic to the underlying LLM, meaning it can work with different models without requiring deep integration. This flexibility allows the system to be updated or swapped as LLM technology evolves, avoiding vendor lock-in. The authors—Liliana Ardissono, Fabio Ferrero, Angelo Geninatti Cossatin, Claudio Mattutino, and Noemi Mauro—present a practical approach to augmenting a heritage guide with LLM capabilities while keeping the core system stable.

Triangolazioni already provides curated information about cultural sites; the new addition enables context-dependent information search and presentation. For example, a visitor might ask a question about a monument and receive an answer tailored to their specific interests or the current visit context. The system's LLM-agnostic nature also allows researchers to test multiple models and select the most appropriate one for the task. The paper, available on arXiv (2608.00549), demonstrates how generative AI can be integrated into domain-specific applications without disrupting existing workflows. This work highlights a growing trend of using LLMs to enhance user experiences in cultural heritage and education, offering a scalable framework for building personalized guides that leverage both curated knowledge and real-time, context-aware generation.