Allegro Tech Talks #48 - Causal Inference AI Edition
Zapraszamy Was na #48 wydarzenie z serii Allegro Tech Talk, podczas których dzielimy się wiedzą, wzajemnie inspirujemy oraz integrujemy przy dobrej kawie ☕, napojach🥤 i pizzy 🍕. Ten meetup odbędzie się w Warszawie, w biurze Allegro i jest częścią serii spotkań, na których rozmawiamy o tematach związanych z AI. Nie zabraknie dobrych historii, praktycznych przykładów i ciekawych insightów. Do zobaczenia 27 maja 2026 roku o 18.00!
Temat wydarzenia: Allegro Tech Talk - Causal Inference AI Edition
*** The event will be held in English ***
Data: 27.05.2026 (środa)
Czas: 18:00 - 21:00
Offline: Warszawa (biuro Allegro - ul. Żelazna 51/53, Norblin Factory office complex) - do zapisanych osób prześlemy wszystkie wskazówki logistyczne
AGENDA
18:00 - 18:10 - Kick-off & a few words about Allegro AI Hub - Kamil Konikiewicz (Director, Data Science - Allegro)
18:10 - 18:30 - Paweł Olszewski (Data Scientist, Allegro) - How are DAGs made?
18:30 - 18:35 - Przerwa / Break
18:35 - 19:35- Aleksander Molak (external guest) - 5 Causal Inference Ideas for the Age of Vibes
19:35 - 19:40 - Przerwa / Break
19:40 - 20:30 - Kamil Golis (Senior Data Scientist, Allegro Pay) - Connecting the Dots: Navigating Credit Risk with Causal Discovery at Allegro Pay
20:30 - 21:00 - Zakończenie & networking / Closing & networking
How are DAGs made?
Formal causal inference often starts with the assumption that the data, its dimensions, and even a full causal DAG are already given. Reality is different, even in seemingly simple settings. Decisions regarding data selection, volume, and structure are made early on and are not primarily driven by causal identification. Causal assumptions easily find critics but rarely have owners. When building a dynamic pricing solution at Allegro we tried to do it by the book and draw a causal DAG. This talk will highlight selected challenges that we faced.
5 Causal Inference Ideas for the Age of Vibes
In the post-big data revolution era, filled with the excitement about "agentic workflows" many businesses face the same existential questions they faced a decade ago: what should I do next, and would it be really better for me to do A compared to doing B? Causal questions underlie any important decision in both business and life. We fantasize about counterfactual scenarios, wondering whether our lives would be better had we chosen a different school, partner, or vacation destination. We calculate potential trajectories a chess match can take, and combine earlier research results with large language model outputs and our gut feelings to create the next marketing campaign or promotional strategy for our employer. In the talk, we'll discuss 5 ideas from causal inference that can help us make better or more informed decisions whenever the stakes are too high to just rely on our intuitions:
- What Are Experiments Really Testing?
- Statistical Significance vs EvidenceGeneralization vs Collider Bias
- No, Double Machine Learning is Not Magic
- Three Ways to Work with Hidden Confounding
Connecting the Dots: Navigating Credit Risk with Causal Discovery at Allegro Pay
In the world of FinTech, predicting what a customer will do is standard practice. But at Allegro Pay, we are diving deeper to understand why. Traditional credit risk models excel at finding correlations, yet they often struggle when the underlying economic environment shifts.In this talk, we explore the transition from purely predictive modeling to Causal Discovery and Bayesian Networks. We will discuss how "connecting the dots" between hidden variables allows us to build more robust, interpretable, and stable risk frameworks.
Paweł Olszewski
Data Scientist with 4.5 years at Allegro. Physicist by education. I'm passionate about statistical inference, uncertainty quantification, and causal inference. I enjoy science fiction and tea.
Aleksander Molak
Aleksander (Alex) Molak is an independent consultant, researcher, and educator specializing in causality who has gained experience working with Fortune 100, Fortune 500, and Inc. 5000 companies across Europe, the USA, and Israel, designing and building large-scale machine learning systems. On a mission to democratize causality for businesses and machine learning practitioners, Alex is a prolific writer, creator, international speaker, and the author of an international best-seller, Causal Inference and Discovery in Python.He's the founder of CausalPython.io - a company that provides training and consulting in causal inference and decision-making for corporate teams, and the host of the Causal Bandits Podcast (https://causalbanditspodcast.com).He's also a part-time tutor in Causal Machine Learning at the University of Oxford.
Social Links:
- LinkedIn: https://www.linkedin.com/in/aleksandermolak/
- YouTube: https://www.youtube.com/@CausalPython
- Book: https://amzn.to/46PperrCausalPython.io: https://causalpython.io/
- Podcast, blog, and more: https://bit.ly/aleksander-molak
Kamil Golis
Credit Risk Data Scientist @Allegro Pay. By day, Kamil builds the credit risk models that keep things running smoothly. By night (or at least in his professional interests), he explores the frontiers of Uncertainty Quantification and Explainable AI to better understand the mechanics of risk. An amateur writer with a taste for "weird stuff," he's on a mission to make sure every technical solution is both transparent and robust.
Do zobaczenia w Warszawie