Imagine building a travel recommender system that truly understands a traveler’s unique needs — from budget and personal interests to walking distance preferences and even local air quality. Sounds great, right? The challenge is: the datasets we have today don’t capture all those nuances, especially for sustainable travel planning.

In this talk, I’ll introduce SynthTRIPs, a project where we use LLMs (Gemini & Llama over VertexAI) grounded in real-world knowledge about European cities to generate realistic user queries — paired with rich personas and explicit travel constraints. These aren’t just random prompts; they’re fact-checked, diverse, and aligned with people’s real travel goals, including sustainability.

I’ll walk you through how we designed the pipeline, the kinds of personas and queries we created, and what we learned from validating them with both humans and AI tools. Beyond tourism, you’ll see how this approach can improve personalization and benchmarking for recommender systems in many domains.

If you’re interested in personalized AI, recommender systems, or just love the idea of smarter, greener travel tech — this session is for you.

Based on our recently accepted paper: Ashmi Banerjee, Adithi Satish, Fitri Nur Aisyah, Wolfgang Wörndl, and Yashar Deldjoo, in Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR ’25), July 2025.

[All the data, code & resources are open-sourced for this project]