PyData Munich

We would like to invite you to our Summer edition of the meetup with three exciting talks on the emerging trends in Generative AI & Vector Search at a very cool central location, for which we want to thank KI performance. We would also like to thank our other sponsor, Couchbase for the food & drinks.

Join us for an evening exploring how the latest advances in GenAI, foundation models, and vector search are reshaping the way we interact with data. From unlocking in-context learning on complex relational databases to transforming vague user queries into precise results, this session brings together cutting-edge techniques that power the next generation of intelligent systems.
We’ll dive into:

• How foundation models can operate directly on structured, multi-table data
• How vector search and GenAI decode user intent for smarter search
• Practical insights into keyword, semantic, vector, and hybrid search strategies

Whether you're working with structured data, building smarter search features, or curious about how AI is accelerating decision-making, this meetup will offer practical knowledge and inspiration from leading experts in the field.

=== Agenda ===
6:00pm Doors Open & Networking
6:30pm Quick Introduction from the Group & Hosts
6:35pm Foundation Models for In-Context Learning on Relational Data by Matthias Fey & Vid Kocijan, Kumo.AI
7:10 pm How to Decipher User Uncertainty with GenAI and Vector Search by Alexander Krasilnikov, Couchbase
7:40 pm Break with Food & Drinks
8:10pm Decoding Search: Understanding Keyword, Semantic, Vector and Hybrid Approach by Gregor Bauer, Couchbase
8:40pm Networking

===Talk Descriptions===

Foundation Models for In-Context Learning on Relational Data
Speakers: Matthias Fey, Founding Engineer @ Kumo.AI and Creator of PyTorch Geometric & Vid Kocijan, Applied ML Engineer @ Kumo.AI

This talk explores how foundation models, originally developed for unstructured data such as text and images, are now enabling in-context learning on structured relational data. We will examine how recent developments allow these models to generalize across diverse tabular prediction tasks without retraining, by leveraging schema-aware
representations and attention mechanisms over multi-table structures. The session will highlight emerging research directions at the intersection of deep learning, graph-based transformer architectures, and multi-modal relational datasets. Throughout the presentation, we will learn how these recent innovations allow an expert practitioner to reduce the time to prediction from months to seconds by introducing predictive models that operate directly on the raw database.

Speaker Bios: Matthias Fey is the creator of PyTorch Geometric (PyG), a leading library for representation learning on graphs. At Kumo.ai, he heads the Research team, driving innovation in foundation models, graph based architectures and scalable ML systems. Prior to this, he completed his PhD on Message Passing for Learning over Graph Structured Data at TU Dortmund University. His work bridges early-stage research with real-world outcomes across a wide range of industry applications.

Vid is a Applied ML Engineer on the Research team at Kumo.ai, where he develops AI-powered products. His work spans foundation models, language processing, user-centric AI systems, and Kumo’s predictive querying language which ultimately enable novel predictive AI solutions for relational data. Prior to joining Kumo, Vid earned his PhD from
the University of Oxford, where he focused on improving the pre-training of neural network models on unstructured data.

***

How to Decipher User Uncertainty with GenAI and Vector Search
Speaker: Alexander Krasilnikov, Solutions Engineer, Couchbase

Unlock the power of generative AI and vector search to transform vague queries into precise results. Discover practical Python examples and see how advanced search revolutionizes user interaction and business outcomes.

***

Decoding Search: Understanding Keyword, Semantic, Vector and Hybrid Approach
Speaker: Gregor Bauer, Manager Solutions Engineering (Central Europe), Couchbase

Learning the different search techniques is essential for developers aiming to implement effective search functionality. In this talk we’ll break down keyword, semantic, vector and hybrid search approaches. We will explore how each method works, their advantages and disadvantages, and practical use cases. This talk is for developers created by a developer and will break down what can be overly complex concepts into practical takeaways for our everyday work. By the end of the session, you’ll have a better understanding of when and how to use each search technique to optimize your user experience.