I like open-source ML tools, and I like them working where the work actually happens. Most of what I’ve built as a Senior Solutions Architect at AWS in Vienna comes from that: getting MLflow and DVC to run properly on Amazon SageMaker, with the access control, lineage, and versioning production demands.
One went further than expected. Open-source MLflow had no access control and enterprises needed it, so I built it on AWS-native services and contributed the client-side SigV4 signing upstream to MLflow. The SageMaker service team picked up the thread: the SigV4 work led to the sagemaker-mlflow plugin that handles authentication in managed MLflow today, and I ended up in the design discussions for what became managed MLflow on SageMaker AI.
These days the same instinct points at agentic AI: my current project is a re-imagined developer portal — describe your application, and an agent bootstraps the infrastructure to best practices.
Build
Production ML with AWS customers — architecture reviews, proofs of concept, and multi-day hackathons. When the blocker turns out to be the tooling, the fix goes upstream.
Experience →Teach
External lecturer at TU Wien on AI & Generative AI, and author of public workshops that AWS field teams run worldwide.
Speaking & Teaching →Write & Speak
8 posts on the AWS ML Blog and talks at AWS Summits, Cloud Days, and meetups across Europe.
Writing →Open Source
Merged pull requests in MLflow (SigV4 authentication, SageMaker container builds) and the SageMaker Python SDK, plus two public workshops delivered to thousands of customers. Every one of them started as a real problem in a customer engagement.
Background
Before AWS I spent four years at RadarServices, a Vienna cybersecurity company, going from developer to team lead to architect on a cloud-based security monitoring product. Before that I was a researcher: a Ph.D. at Trinity College Dublin, where I started out programming Software Defined Radios (SDRs) — you can check that work here. My research focused on mobile network infrastructure sharing between operators, where I analyzed call detail records from two major mobile operators in Ireland, with papers in IEEE Transactions on Computers, Big Data, and Networking. The common thread, in hindsight, is distributed systems and data — I’ve just kept moving up the stack.
Get in touch
Want to talk MLOps, agentic AI, or invite me to speak at your event? The best way to reach me is LinkedIn; for anything open-source, open an issue or find me on GitHub.