Team lead of the working group "traffic systems" in the department "Nautical Systems" (Institute of Communications and Navigation). My team is working on our cutting edge solutions for analyzing and predicting the maritime traffic network, spotting anomalies as well as target tracking and shape estimation of vessels using maritime X-band radar.

As a research engineer, I conducted research on predictive uncertainty estimation with neural networks and speeding-up CFD simulations with physics-informed neural networks and (Fourier-based) neural operators. Furthermore, I developed efficient C and C++-based implementations of algorithms in the realm of Reinforcement Learning.

After participating as an AI researcher in FDL's 2021 research sprint, I became a team lead of FDL Europe's 2022 Aersol team. Together, we sought to use ML techniques to improve the detection and our understanding of Pyrocumulonimbus (pyroCb) clouds. These wildfire-generated weather systems can lead to more lightning strikes, igniting even more fires in the landscape. They are becoming more common globally, driven by climate change, and we need to understand when, where, and how they form.Assembling a catalog of pyroCb events is challenging because they are still relatively rare with only a few hundred examples in years of satellite data. Leveraging existing records, we trained a machine learning model to find pyroCb clouds in over 400 terabytes of data from the Himawari-8 and GOES-17 geostationary satellites.Using our new catalog, we have been digging into the major factors that predict if a wildfire will develop into a pyroCb event and how we might mitigate them in the future.

Anomaly detection in safety-critical maritime traffic movements using non-linear Kalman Filters and AI/ML techniques at the Institut of Communication and Navigation (Nautical Systems). (Details are subject to NDA.)

I joined the collaboration between NASA Ames, the SETI Institute, and commercial AI partners like Google Cloud to minimize the deleterious effects of climate change. Our four-person team's approach to this problem was to speed up flood predictions by building a machine learning surrogate model for an established inundation height model like NEMO (Nucleus for European Modelling of the Ocean). We were the first to apply the physics-informed state-of-the-art Fourier Neural Operator models to a real-world setting and predicted inundation height better than the physical and ML baselines. Our work is a step in the direction of making practical, high-precision flooding predictions a possibility, which would inform evacuation orders and save lives.

In the large data sample of the LHCb experiment at CERN I searched for rare b to open-charm two-body decays using physical intuition and statistical / ML methods. The results of this study are the first discovery of a decay by observing it with a statistical significance above five standard deviations, and estimations of upper limits of another, back then not yet observed, decay.During my PhD time I had the fantastic opportunity to visit and participate at various international HEP, HPC and ML orientated schools and conferences. These thrilling meetings boosted my passion for high performance computing and AI research and granted insights into leading edge developments.

Internationale Summer School that offers students in physics or related natural sciences to participate in DESY’s research activities. About 100 students from all over the world attended a rich lecture program and worked on small projects. As part of this work, I implemented a C++ program that allows the creation of drivers for Slow Control Producers that are dynamically linked to the EUDAQ framework during runtime of the program. In comparison with the previous solution this reduced the number of time consuming compilation steps and thus significantly speed up the daily laboratory workflow.