Experience

I build core software infrastructure for the foundation model research team, enabling researchers to develop AI that enhances personalized recommendations for our nearly 300 million users worldwide.

I deploy and maintain ML solutions across 70+ countries at the global leader in chemical distribution. Recently, I migrated Brenntag’s flagship in-house virtual AI assistant, which earns €30+ million in annual revenue, to a more cost-effective and secure AWS platform. Outside of infrastructure engineering, my primary focus is increasing ML product scalability.

As a contracted ML engineer and AI developer, I optimized and implemented various ML solutions at tech startups, scale ups, and large tech companies. Notably, I spent 1 year at a leading e-classified ads platform working in an experienced team (8+ years of development experience). During my tenure, I increased the stability of the “For You” homepage recommendation system.

During my master’s thesis, I researched solutions to a key obstacle faced by people analytics platforms like Crunchr: encoding highly heterogeneous data (source paper). This meant encoding relational database data in a scalable way that most ML algorithms could “understand". I delivered a one-off, graph-based representation learning process. I validated my solution by using deep neural networks to classify the outputs of my process in downstream tasks.

I established the technical foundation for automating cardiovascular disease risk prediction in a clinical research setting. According to the literature, excessive epicardial fat is associated with an increased risk in heart failure and other cardiovascular diseases. Hence, I built an end-to-end neural image segmentation pipeline to automate epicardial fat volume measurements in patient CT scans. My deliverable served as a tool for subsequent medical research.

At Cubelizer, a Google-backed startup using computer vision for retail space optimization, I improved the accuracy of base customer detection by 12%. I developed a real-time object detection method, utilizing video processing and classical computer vision techniques. My solution fully complied with EU privacy regulations.