AI-Native Platforms: From AI-Assisted Operations to Running AI Workloads

THIS MEETUP WILL BE IN PERSON 🕺

What to expect:

• 🗣️ Two or Three speakers
• 👥 Meet old and new friends from the community
• 🍕 Free food
• 🎉 Open-end and free drinks afterwards
• ✔️ Just RSVP!

Agenda:

• 17:30 - Doors open
• 18:00 - Welcome
• 18:05 - 🗣️ Dmytro Hlotenko - My Step Functions DAG Was a Lie: Going Agentic on 8,000 Photos
• 18:50 - Break
• 19:00 - 🗣️ Martin Flechl - From Alert to Fix: Agentic Root Cause Analysis and Self-Remediation
• 19:40 - 🗣️ Marco Basmaji - LLM Observability: What Happens When You Can't Just Read the Stack Trace
• 20:00 - 🎉 Open-end

🧑‍💻 Speakers:
Dmytro Hlotenko - Senior Cloud Operations Engineer at Sage Austria
Dmytro Hlotenko is a Senior Cloud Operations Engineer at Sage Austria, working on the cloud platform behind its HR and payroll products for the DACH market - the kind of environment where decades-old business logic meets a modern hybrid cloud estate, and where reliability is not negotiable.
He is an AWS Community Builder, co-lead of AWS User Group Vienna, an AWS New Voices 2025 graduate, and an event manager with AWS Community DACH e.V., where he helps organise AWS Community Day DACH and co-organised AWS Community GameDay Europe in Vienna. Seven AWS certifications later, he still believes the interesting part of the job is the part that does not fit the reference architecture.
He has spoken at AWS Community Days across Europe, at Conf42, on the re Recap Tour, and at the AI for Developing Countries Forum at the United Nations in Vienna. Ukrainian, based in Vienna, and an analog photographer — which is how the agentic photo pipeline he now talks about got started in the first place.

Dmytro's session
My Step Functions DAG Was a Lie: Going Agentic on 8,000 Photos
Once I decided to translate my photography experience into a serverless pipeline. I judged 8,000 photos of your faces, and it was a big success - we turned a problem into a solution with AWS. Lambdas, Rekognition, Bedrock and the rest played together as a system, delivering a memorable AWS Community Day.
The talk traveled far - but it had a fatal weakness. It was good while it lived in the conference room. Outside, my taste — hardcoded into weights and thresholds, and a single Claude call guessing how to process every file — brought things to disaster. It ran rules I wrote for photographs I hadn't taken yet, and failed silently on every shot that didn't fit. Terraform destroy.
This year I tore out every guess. With agents.
Imagine swapping the single photographer for the press agency. I'll show how to keep a mature orchestration relevant in the agentic era. How a chain of pipelines became a grid of agents - separate programs, each with its own tooling and skills distilled from real photography craft. Where the agent reasons about what belongs and judges against real craft, not one engineer's gut. Where v1 had a frozen rule, v2 has a program that thinks.
I'll draw the parallels — where we broke through, where classic approaches still win, which weaknesses got replaced and what each agent does instead. You'll leave with a pattern for any rigid pipeline: which box is a guess, and what if it could reason instead?
Last year I automated my hands. This year, my judgment.

Martin Flechl - Director of Generative AI Research at Dynatrace
Martin Flechl is a machine learning specialist, former particle physicist, and Director of Generative AI Research at Dynatrace, where his team works on LLM- and agent-based approaches to observability and automation.
Before joining Dynatrace, Martin worked at Microsoft on speech recognition and language modeling. Earlier in his career, he spent many years working in particle physics and large-scale scientific data analysis — including as part of the international research effort that led to the discovery of the Higgs boson.
With more than 20 years of experience in data science, data engineering, data analysis, and machine learning, Martin has spent much of his career turning enormous amounts of complex data into useful insights. He has programmed in Python and C++ for more than two decades and brings extensive experience leading research teams and managing international scientific projects.
He has also taught at university level for more than 20 years, combining deep technical expertise with a strong interest in explaining complex ideas clearly.
Today, that combination of scientific data analysis, machine learning, research leadership, and AI is focused on a very practical problem: what happens when software breaks in production, and how far can AI go in understanding the problem and fixing it?

Martin's session
From Alert to Fix: Agentic Root Cause Analysis and Self-Remediation
AI coding agents are increasing the rate at which code reaches production. And with it, the rate at which production breaks. The bottleneck moves from writing code to understanding and fixing what it does at runtime.
This talk looks at what it actually takes to close that loop. Martin will start with root cause analysis that combines topology, traces, and logs with LLM-based reasoning, then show how that analysis turns into something an agent can act on: the specific commit and code path at fault, the blast radius, and an evidence-backed issue handed to a coding agent.
He'll also cover the part that gets less attention: measuring whether any of this actually works.

Marco Basmaji - DevOps and Platform Engineer at SQUER
Marco is a DevOps and Platform Engineer with extensive experience across projects of every scale, from targeted automation efforts to large, complex infrastructure builds. His recent focus is on AI enablement, bringing intelligent automation into scalable, reliable systems.

Marco's session
LLM Observability: What Happens When You Can't Just Read the Stack Trace
Traditional observability was built for deterministic systems: a request comes in, a predictable path executes, and when something breaks you get a stack trace. LLM-powered systems break that model — the same input can produce a different output, failures look like plausible-sounding nonsense instead of exceptions, and cost, latency, and quality are now entangled in ways that classic APM was never designed to capture.

📍 Location
SQUER Solutions
Althanstraße 4/3/63
1090 Wien, Austria

📢 Got something to say?
Are you a seasoned speaker, or is this the first time you hit the stage?
Want to give a talk at one of our next events? We would like to have you.

Our formats:

• Full talks (~30 to 45 min.)
• Short talks (~15 min.)