On 19 November I will be speaking at Google Berlin's AI Innovation Hub.
The occasion is the German Data Science Society's autumn conference — the Herbsttagung — a single-track day that brings together practitioners from industry, startups and research. The theme this year is what it actually takes to put data science, GenAI and AI agents into production, which is squarely the problem we work on.
What the talk is about
Predictive modelling is becoming more accessible. Tabular foundation models, time-series foundation models and a new wave of decision models are making powerful predictive capabilities available with far less model-specific work than a few years ago.
Yet turning a business question into a useful predictive system still takes substantial effort. It means framing the problem, understanding how enterprise tables fit together, assembling the right data and defining the target. It also means preventing leakage, designing realistic validation, choosing meaningful baselines and metrics, and connecting predictions to decisions and operational workflows.
Much of applied data science lives in these steps — not in the model. Agentic workflows offer a way to automate more of them, with agents coordinating data preparation, classical machine learning and specialised models.
Drawing on the experience of building Predictabl, I will share practical lessons from automating this work: where agents genuinely help, where they fail, how to recover when they do, and where human judgment is still essential.
The argument I want to make is that automating the work around the model will reshape the data scientist's role more profoundly than improvements in predictive models alone.
Details
- When — Thursday 19 November 2026, 9:00–18:00
- Where — Google Berlin, AI Innovation Hub, Tucholskystraße 2
- Format — one track, all day, most talks in English
- Tickets — via the German Data Science Society
If you are going to be there, come and say hello.