74th Deep Learning Meetup: Agentic AI & Causal Inference

Hi Deep Learners,

We are happy to announce one more Vienna Deep Learning Meetup before the summer break: on June 10 at ÖBB. We will again have two talks: Agentic AI in Production and Exploratory Causal Inference.

***
Agenda:

• 18:15 Arrival
• 18:30 Introduction by the meetup organizers
• Welcome by the host: ÖBB
• 18:45 Talk 1: Agentic AI Systems in Production: Best Practices, Challenges & Lessons Learned by Hilda Kosorus (Onefold AI)
• 19:30 Announcements
• Networking Break
• 20:00 Talk 2: Scaling Empiricism in Artificial Causal Inference by Riccardo Cadei (ISTA)
• 20:30 Networking
• ~21:30 Wrap up & End

***

Talk Details:
--------------
Talk 1: Agentic AI Systems in Production: Best Practices, Challenges & Lessons Learned

Building agentic AI systems that work in production is harder than the demos suggest. Real users, real data, and real edge cases surface challenges that don't show up in most tutorials. We must consider orchestration trade-offs, observability gaps, evaluation strategy, latency and cost realities. There's constant tension between giving agents autonomy and keeping them under control.
In this talk, we share what we've learned shipping agentic systems end-to-end. We'll walk through how our architectures evolved and the trade-offs behind each shift, and we'll be honest about which frameworks — such as LangGraph, LangChain, Langfuse — earned their place in our stack. Expect architecture diagrams and the patterns we now apply to lead our custom projects to success.

About the speaker:
We are Onefold AI — Tobi, Csenge, and Hilda. We combine deep LLM engineering with a research foundation, years of data science work, and multiple agentic systems shipped end-to-end into production. Our recent projects span compliance and audit automation, multi-agent workspaces for food scientists, and our own product development — all building on the hard-won lessons we'll share in this session.

Talk 2: Scaling Empiricism in Artificial Causal Inference
Randomized trials are the gold standard of empirical science, yet their analysis still hinges on hand-crafted hypotheses: the investigator has to decide upfront what to measure and whom to compare, often anchoring on familiar narratives. A paradigm shift is now within reach: modern trials measure more, and representation learning gives us the tools to scale the reading accordingly. In this talk, I will present two algorithms I developed to bridge this richer measurement to causal claims, e.g., scientific discoveries or policy guidelines. Neural Effect Search (NES) identifies the latent effects of a treatment from unstructured outcomes. Neural EXposure Interaction Search (NEXIS) identifies an interpretable and prescriptive characterization of effect heterogeneity. I will illustrate both through real-world deployments in experimental ecology and development economics.

About the speaker:
Riccardo Cadei is an ELLIS PhD student at ISTA with a growing record of foundational, methodological, and applied contributions to Causal Inference and AI for Science. He actively collaborates with biologists, neuroscientists, economists, and public-health researchers, translating his vision of Artificial Causal Inference into concrete scientific workflows.

We are looking forward to welcoming you at our last meetup before summer!
Your VDLM organizer team