#3 Data Horizons

You’ve probably heard the buzz around Data Mesh – and maybe even the growing skepticism, with some calling it passé due to the implementation challenges. But don’t write it off just yet as there’s still plenty of ground to cover before the next wave of data architecture trends kick in.

In this talk, we’ll take a pragmatic look at what domain-driven data architecture really looks like in practice. We’ll walk through the key concepts behind Data Mesh, the platform design choices that shape its implementation, and the realities of putting it all into action. From Terraform struggles to the argument for domain-specific languages, to the chaos of scaling self-serve data – this is part reflection, part roadmap, and rooted in real-world experience.

Everyone wants AI. But often what they really have is: no clean data, no dedicated team, no real budget, and no idea who’s supposed to use the model once it’s done.

In this talk, I’ll share what happens when you build AI in environments full of chaos, shortcuts, and wishful thinking — and still aim to deliver something useful. You’ll hear lessons from real-world deployments: predicting churn without stable data pipelines, building scoring models no one wants to own, and negotiating delivery timelines with seven stakeholders and zero documentation.

This is a talk about pragmatic, compromised, and actually used AI – not the kind that looks great in a notebook and dies in production.

One of the fundamental features of LLMs is their ability to provide responses aligned with human ethical principles and intentions. In his talk, Przemek will explore what LLMs truly “know” about morals, discuss how models are being trained for AI Alignment, and reveal how (surprisingly) easy it is to break. He’ll show experimental results, in which a small set of harmful examples could derail alignment. A live demo and open discussion on what it all means for us will follow.

For all its convenience, batch processing is far from a perfect solution. In this lightning talk we’ll highlight the major, most common data issues when working with classic batch processing and ETLs, while providing insight on how to resolve them with stream processing.