Professional Experience

My day job is helping customers take machine learning to production — startups, enterprises, public sector, and universities. Most engagements are about MLOps, DevOps, or generative AI: architecture reviews, proofs of concept, and the occasional rescue of a project that grew faster than its foundations.

As of 2026 I've written 8 posts for the AWS Machine Learning Blog. My early MLflow-on-AWS work — customer POCs, blog posts, and an upstream authentication contribution — led to design discussions with the AWS service team during the development of managed MLflow on Amazon SageMaker AI, where I brought the customer perspective and lessons from running MLflow on AWS in the field. When customer problems turn out to be tooling problems, I fix them upstream — merged contributions to MLflow and the SageMaker Python SDK (details on the Open Source page).

I wrote two public workshops that AWS field teams deliver worldwide: the Amazon SageMaker Immersion Day and MLOps: from idea to production in six steps. Thousands of customers have gone through them.

My main project in the agentic AI space re-imagines the developer portal experience: instead of clicking through a portal, developers describe their application and an agent bootstraps the infrastructure across AWS accounts according to codified best practices. Built on Strands Agents SOPs — machine-readable runbooks with RFC 2119 keywords, parameterized inputs, and progress tracking — the reference implementation takes a web application from description to production across three AWS accounts (~150 resources, 6 Terraform modules) in under an hour, with a human reviewing the plan and gating production. The same task used to be a two-week runbook exercise. I've presented this work several times, and it's where my current thinking on agentic systems comes from.

I run multi-day hackathons for enterprise customers; teams leave with a working prototype, and several have taken theirs to production. I speak regularly at AWS Summits, Cloud Days, and meetups across Europe (see Speaking & Teaching), mentor new Solutions Architects, and help build the AI/ML community in Austria.

My generalist years. I helped customers from startups to enterprises adopt AWS: cloud migrations, multi-account landing zones, cost optimization, and a lot of whiteboard sessions across manufacturing, media, energy, and technology. This is where I learned that the most valuable thing an architect brings to a review is not answers, but better questions.

I led the transformation of a large monolith into a service-oriented architecture for a cloud-based IT security monitoring product. That meant designing and documenting the REST APIs (data formats, versioning strategy), hunting down the performance bottlenecks that threatened our SLAs, and sitting with enterprise customers to turn their business processes into requirements the team could build. I also mentored the junior developers who would later inherit the system.

Led a development team through delivery, hiring, and the daily work of keeping stakeholders and engineers pointed in the same direction.

Joined as a developer, shipping features in a test-driven environment with peer review. Nine months later I was leading the team.

Worked on the Fed4FIRE EU FP7 project, building the frontend and backend that gave researchers remote access to Trinity's wireless testbed, including a web-based resource management system with automated assignment.

Built and maintained nationwide datasets of anonymized call detail records from O2's cellular network, then analyzed the spatio-temporal dynamics of traffic demand. That work became the empirical backbone of my network planning research — and my first real lesson in how much of data science is data janitorial work.

Taught Java programming to Computer Science students and ran the Electrotechnics lab — filtering, modulation, and hands-on oscilloscope work.

Built a split architecture for carrier-sensing MAC protocols on Software Defined Radio platforms, and gathered the experimental data to evaluate them in multi-hop wireless networks.