From Prompts to Pre-Training: Quality, Scale, and the Road to AI Native

We are thrilled to welcome you to our 10th edition of the AI Native Netherlands, hosted at the Adyen headquarters in Amsterdam!

This is a joint meetup between Adyen Developer Events and Meetups and AI Native Netherlands meetup.

At the end of the evening, we're adding another Q&A panel. After the talks, you can submit questions online about what was just presented. The speakers will answer them live.

A massive thank you to our host, Adyen, for their generosity in providing the venue, food, and drinks for this evening and for supporting the Dutch AI engineering community.

We’ll cover:

• Multi-layer evaluators and human-in-the-loop safeguards for production LLM systems
• Failure modes and silent quality degradation in agentic pipelines
• Quality engineering and monitoring for production LLM systems
• Building and deploying foundational models at trillion-token scale
• Training infrastructure, pre-training at trillion-token scale, and meeting strict production SLAs
• A structured panel discussion where your questions about the talks drive the technical deep dive

Speaker 1: Ronny Roeller (NEXT AI)
Ronny is CTO and Co-Founder of NEXT AI — a customer insights platform for the AI era. Launched in April 2023, clients include Deel, Bosch, Generali, Rituals, Action, and others. He studied Computer Science at VU Amsterdam and works on building scalable, production-ready AI systems.

Talk: From Prompts to Production: Engineering Quality in LLM-Based Systems
Building systems around large language models is fundamentally different from traditional software. Outputs are probabilistic, data-dependent, and costly to compute. Deterministic QA approaches quickly break down.

In this talk, Ronny will share how production-grade quality was engineered for LLM workflows using multi-layer evaluators, real customer datasets, and human-in-the-loop safeguards. He will walk through concrete failure modes encountered in production, how silent quality degradation emerged over time, and the architectural patterns that restored reliability.

The focus is on practical techniques to measure, monitor, and continuously improve LLM system quality beyond prompt tuning.

Speaker 2: Hanna van der Vlis (Adyen)
Hanna van der Vlis is an AI research engineer at Adyen, a payment processor that handles over 40B transactions annually. She focuses on deep learning research, training infrastructure, and building foundational models. She is passionate about solving large-scale engineering challenges, deploying AI to optimize complex systems and driving real-world impact.

Talk: Towards The Adyen Foundational Model
Traditional machine learning in the payments industry relies heavily on point-in-time information silos and Boosted Trees. This approach is increasingly constrained by manual feature engineering and limited label availability.

In this talk, Hanna will outline Adyen’s transition toward a unified Foundational Model by rethinking payment sequences as a language problem. She will walk through the engineering challenges of pre-training a shared backbone on up to 51.2 trillion tokens, overcoming compute bottlenecks, and fine-tuning for downstream tasks such as fraud detection and authentication.

The focus is on the practical realities of developing and deploying these large-scale models while meeting strict sub-30ms SLAs in production.

Agenda:
17:45 — Arrival, food & drinks
18:30 — Talk #1 | Ronny Roeller
19:15 — Talk #2 | Hanna van der Vlis
20:00 — Panel discussion: Open conversation & audience Q&A
20:30 — Networking & more drinks
20:45 — Wrapping up

What to bring: Just curiosity and questions. If you are working on LLM workflows, foundational models, training infrastructure, or sub-30ms production SLAs, we would love to hear your technical questions during the panel discussion.

Who is this for: Platform engineers, AI/ML engineers, SREs, architects, and engineering leaders focused on building and validating reliable AI systems in production.

Where to find us: Adyen, Simon Carmiggeltstraat 6-50, Amsterdam
(meet at the reception; you will be brought to the 11th floor of the office building where the event will take place).