Sésame

Sésame is an AI-powered personalized guide generator, integrated directly into the Tourisme Montréal website. Not a chatbot, not an enhanced search bar. A complete and unique guide, generated in seconds: a personalized image, a title, a summary, recommendation texts tailored to the query, editorial articles and events tied to the user's context, as well as all external context (business opening hours, time of day, season, weather). Sésame is also a virtuous lens: other visitors' guides are visible, and in turn become an invitation to explore the city differently.

To achieve this, the technical architecture could not rely on anything fabricated. Every recommendation had to be grounded in the platform's human-generated content : no hallucination, no generalities, a true agentic RAG. An agent first enriches the user's vague intent. Context engineering then takes over: real weather, time of day, ongoing events, and opening hours are assembled before each generation so that the AI always reasons within the reality of the moment. Two engines work in parallel: one for semantic proximity (dense retrieval), the other for factual precision (sparse retrieval, BM25). A diversification algorithm (MMR) prevents the same names from always surfacing. A contextual reranker reorders results based on the full picture : event timing, weather, season, time of day, and interests extracted from the original query. That is why Sésame won't send you back to the same museum for the tenth time when a pyramids exhibition is in town and you came with friends who share your passion for history. Because today, for you, at this moment, that is what is relevant.