Wed, Jan 21 · 6:30 PM CET
Welcome to the PyData Berlin January meetup!Make sure to arrive on time! Host :Spiced Academy is excited to welcome you to this month's version of PyData.Talk 1: From Grape Stomping to Gorilla Trekking: AI-Driven Discovery of Novel Interests at GetYourGuide Abstract : UX research shows that discovering new, unexpected and unique things to do in a destination is among the most important needs for travelers. However, surfacing these “hidden gems” at scale is a major challenge for any dynamic marketplace. At GetYourGuide, we tackled this by combining large language models, semantic vector embeddings, and a robust human-in-the-loop process to automatically identify and categorize novel customer interests. Join me as I dive into how we created this AI-driven solution and how we achieved scalable, granular and relevant categorization of our activities.Description: How do you help a traveler find something they never knew they wanted? In this talk, I will explore the exciting journey of developing an AI-driven solution to identify novel customer interests at GetYourGuide. With thousands of new activities onboarded weekly, manual categorization became impractical, prompting the need for automation.
The problem: the challenges travelers face in discovering relevant and unique experiences, the organizational complexity of this on our platform and its business impact
The AI journey: From raw LLM-based extraction to vector embeddings, clustering, all the way to a robust human-in-the-loop pipeline. I’ll share how we iterated from noisy prototypes to an impactful system, launching 300+ new interests and driving significant engagement and net revenue uplifts
What we learned: The realities of working with LLMs in production-like settings, more specifically hallucinations and issues with relevance, granularity and semantic deduplication. We found out that the line between useful innovation and unwanted noise can be surprisingly fine when using LLM’s. I’ll share practical examples of these challenges and how we’re iterating to address them.
Impact and future directions: How we achieved a 10x increase in manual acceptance of AI-suggested interests in just two iterations, and what’s next for AI-driven discovery at GetYourGuide
Bio : Morena Bastiaansen holds a master’s degree in Econometrics from the University of Amsterdam and currently works as a Data Scientist at GetYourGuide in Berlin. She combines her passions for machine learning, language, and travel to build impactful systems for travelers worldwide, with a current focus on personalization in email marketing. Morena is also actively involved in the local tech and ML community.Talk 2: Deep dive into data streaming security Abstract : Data streaming is powering everything from fraud detection and real-time analytics to patient monitoring and order fulfillment. But as the role of streaming grows, so does the risk - because many streaming platforms, like Apache Kafka, aren't secure by default. In this talk, we’ll take a practical look at data streaming security through the lens of Kafka, one of the most widely adopted streaming platforms in the world. We'll walk through what can go wrong - real examples of exposing 14M patient-doctor messages, or years of real-time delivery info to the internet - and what it takes to do it right. We'll cover the key pillars of securing a streaming system: encryption in transit and at rest, access control, monitoring, and key management. Along the way, we’ll look at the trade-offs like disk encryption vs. end-to-end encryption, what is behind field-level and envelope encryption, and the realities of using customer-managed keys in regulated industries. We'll also explore how streaming security has evolved, how real vulnerabilities (like CVE-2019-12399) highlight the need for patching and monitoring, and what successful multi-layered security looks like in production - from financial institutions to healthcare platforms. If you're building, running, or scaling streaming systems, this talk will help you see the security blind spots and give you concrete steps to protect the data flowing through your pipelines.Bio : Olena Kutsenko is a Staff Developer Advocate at Confluent and a recognized expert in data streaming and analytics. With two decades of experience in software engineering, she has built mission-critical applications, led high-performing teams, and driven large-scale technology adoption at industry leaders like Nokia, HERE Technologies, AWS, and Aiven.Lightning talks Code of Conduct
