The Psychology of Data Transformations & Optimizing Multimodal AI Pipelines

Dear Data Enthusiasts,

Join us again on 29th of May at A1 to learn about the psychological implications on employees when transforming the data landscape of an organization. In addition, we will finally have a talk about the open source framework Metaxy and how it can help to cut costs of multimodal AI pipelines.

A big thanks goes to our co-organizers from the A1 Telekom and our sponsor Cloudera!

A Psychological Autopsy of a Data Transformation
Dávid Farkas
Five years ago, a small European university decided to build and implement a data strategy from scratch. Greenfield in the truest sense: data governance was an unheard term and in practice it meant local Excel files duplicated across inboxes; "the database" usually referred to whichever spreadsheet someone had emailed most recently. Over four years, we designed and rolled out new systems, established governance structures, hired a team, and delivered real wins. We also watched a significant portion of the work get undone in months once leadership changed.
This talk is a psychological autopsy of that transformation — what was planned, what worked, what quietly collapsed, and why. Universities are an unusually honest laboratory for studying data initiatives: decentralized power, competing stakeholder logics, weak formal hierarchies, and ambiguous success metrics make every human dysfunction that exists in corporate environments more visible.
Drawing on behavioural science and nearly five years leading this initiative, I'll walk through the recurring failure patterns I see across data and AI projects more broadly — algorithm aversion, NIH syndrome, the gap between executive sponsorship and operational ownership, and the under-recognised role of organisational identity in determining which systems survive a leadership change. The argument is not that technology doesn't matter — it's that the human substrate determines whether any of it sticks. Borrowing Kranzberg: technology is never neutral, and neither are the people implementing it.
Dávid Farkas is a research psychologist (PhD) and data scientist based in Budapest. He co-founded Principle Zero, which helps companies, NGOs, and researchers take on complex challenges in data strategy, digital and AI transformation, research, product development, and science communication. We bring behavioural science to problems most consultancies treat as purely technical, focusing on the human and organisational factors that determine whether ambitious projects survive contact with reality. Previously he led data science and digitalisation at MOME Budapest, with earlier industry experience applying machine learning.

Optimizing Multimodal AI Pipelines with Metaxy
Georg Heiler, Hernan Picatto
The AI era has caused a fundamental shift in computing, moving us toward complex multimodal pipelines. However, these new systems are often incredibly wasteful. Right now, small changes to an input can trigger massive recomputations across very expensive processing steps.
In this talk, we will explore Metaxy, an open source Python framework built to solve this exact problem. Metaxy provides sample level metadata versioning and acts as the universal glue for incremental data pipelines. We will discuss how its field level provenance allows your pipeline to only recompute what actually changed. If you update an audio file, for example, Metaxy knows to skip downstream face recognition steps that only rely on video.
Whether you are a startup stretching your compute budget, an enterprise scaling ML infrastructure, or a researcher in academia, you will learn how to use Metaxy to iterate faster, cut cloud costs, and build highly efficient AI workflows.
Georg Heiler is a co-founder @Jubust and a Senior data expert at Magenta as well as a ML-ops engineer at ASCII. He is solving challenges with data. His interests include geospatial graphs and time series. Georg transitions the data platform of Magenta to the cloud and is handling large scale multi-modal ML-ops challenges at ASCII.
Hernan Picatto is a Computer Science PhD candidate at the Vienna University of Technology researching firm-level supply chains and corporate networks using NLP and Common Crawl data. Formerly an engineer at JPMorgan Chase and ZhiZhouKeji, he holds an MA from UCSD. His broad interests span big data, visualization, and time series causality.

🎤🎤 Open Mic
We are going to open up the stage after the talks for community announcements. If you'd like to announce something, open this slide deck, make sure you are signed in with a google account, and click "View Only" -> "Request Edit Access". Explain in the text box what you want to announce, and we'll give you edit access to the slide deck.
🎤🎤

We’ll have some food and drinks after the event.
Please note that during the event, photos might be made and later posted on VDSG's social media page. Please notify us if you do not agree.
Attention attendees with food allergies. Please be aware that the food and drinks provided may contain or come into contact with common allergens, such as dairy, eggs, wheat, soybeans, tree nuts, peanuts, fish, shellfish, or wheat.

Best,
The Organizer Team