AI in Production: Building Recommenders and AI-Powered Discovery
💡 Before registering for this event, please read the note about event recording and photography at the end of the event details.
👉 Want to stay connected with us and get more insights about GetYourGuide? Check out our Linktree for all the essentials: dive into our tech blog, explore career opportunities, and respond to our speaker call for future events.
Welcome to the 8th AI in Production Meetup 🚀
Curious how cutting-edge AI systems move from research papers to real-world impact? Join us for an evening of practical insights into how modern AI powers large-scale recommendation and discovery systems in production.
In this edition, we’ll explore two very different AI problems. First, we’ll dive into solving extreme item cold-start in dynamic marketplaces, and how content-aware transformer-based recommender systems can generalize to entirely new items in real time. Then, we’ll shift to the world of travel discovery and unpack how an AI agent can turn natural language queries into relevant, ranked experiences and enable travelers to find unforgettable experiences.
You’ll hear lessons learned from real deployments, including scaling strategies, evaluation frameworks, latency–cost trade-offs, and how to move from impressive demos to reliable systems that work at scale.
This meetup is geared toward ML engineers, data scientists, and AI practitioners building production systems—but anyone passionate about applied AI is more than welcome. Come connect with fellow builders, enjoy pizza and drinks, and learn from teams shipping AI products used by millions.
We’ll also open the stage for a few spontaneous lightning talks—so if you’re working on something exciting, experimenting with a new idea, or have lessons learned to share, bring it along and take the mic for a few minutes! ⚡
Talk #1: Solving Item Cold-Start at Scale: Real-Time Recommendations Beyond ID-Based Transformers
Jan Malte Lichtenberg, Principal Applied Scientist @ Albatross AI
Item catalogs are becoming increasingly dynamic, with thousands of often user-generated items entering the catalog every minute, especially in domains like second-hand marketplaces or short-video platforms. This leads to an extreme form of the item cold-start problem, which challenges not only simple popularity-based approaches but also modern ID-based sequential recommender systems built on transformer architectures.
In this talk, I explore strategies to address this problem by incorporating item content information, such as textual descriptions and images, into recommender systems. I begin with a short tutorial to sequential recommender systems and show how they relate to the transformer architecture that also underlies models like ChatGPT, before discussing practical approaches for leveraging content to improve generalization to unseen items.
The talk is grounded in a real-world deployment, including a case study from a collaboration between Albatross AI and the European second-hand marketplace Wallapop.
Talk #2: From Query to Discovery: Building an AI Agent That Helps Travelers Explore
Mathieu Bastian, Principial AI Engineer @ GetYourGuide
Giampaolo Casolla, Senior Data Scientist @ GetYourGuide
Search works great when you know what you're looking for. But what happens when a traveler types "sunset tours in Barcelona for a couple under 50 euros"? Traditional search pipelines break down.
We built an AI-powered discovery agent at GetYourGuide that lets travelers describe what they want in natural language and get relevant, ranked activity recommendations. In this talk, we'll walk through the production architecture: a multi-stage LangGraph pipeline that chains intent interpretation, parallel retrieval, LLM-based relevance filtering, and natural language synthesis. We'll cover the practical trade-offs we made along the way, including a dual-LLM tier strategy (fast nano model for high-volume filtering, full model for nuanced interpretation) that keeps latency and cost in check without sacrificing quality.
We'll also share how we built an evaluation infrastructure with 800+ deterministic tests and no LLM-as-judge in CI, using a pass@k approach to handle output variability while keeping the test suite fast and reliable. If you're shipping LLM-powered features to production and struggling with the gap between "cool demo" and "actually works at scale," this talk is for you.
Schedule of the Meetup:
18:00 Doors open + Open Networking
*Please note that doors will be closing at 18:50.18:30 Introduction to GetYourGuide
18:40 Talk # 1: Solving Item Cold-Start at Scale: Real-Time Recommendations Beyond ID-Based Transformers
19:10 Break
19:30 Lightning Talk Session
19:40 Talk # 2: From Query to Discovery: Building an AI Agent That Helps Travelers Explore
20:15 Raffle (5 min) + Open Networking
21:00 End
ℹ️ The meetup begins at 18:00. Doors will remain open until 18:50, after which entry won’t be possible. We appreciate your understanding and look forward to welcoming you!
Answers for the curious:
Who should join the meetup?
This meetup is dedicated to anyone who works with, studies or is interested in ML and AI. It's a fun and informal environment where we can learn from each other's experiences.
What's the deadline to register for the event?
Please register by 18:00, November 12th.
Can I bring a plus one?
Sure, that's totally fine. Just ensure to register them as well.
Can I bring a dog?
Unfortunately, we don't allow dogs at this event.
For any questions, please don't hesitate to contact us.
Important note: Please be advised that this event will be recorded and photographed, and we will have a photographer on-site. If you prefer not to be included in any recordings or photographs, please do not hesitate to let us know during the event. Your comfort and privacy are important to us.