My Research Profile

An overview of my research agenda and Examples

I strongly value good team dynamics and have developed skills to manage productive meetings, group conflicts, and clear, goal-oriented communication. Seven bachelor, master, and PhD students were successfully mentored by me on their own research projects.

Research Agenda:


Conceptual Research

I think on a conceptual side I would focus on two questions:

  1. How does logical reasoning emerge from correlative reasoning?

Corrlative Reasoning: Up to certain amounts of data and compute, LLMs reason correlatively for a lot of tasks. They associate certain words and concepts with each other and predict next words from sets of words that are highly correlated with previous words. Logical Reasoning: Given enough data, the LLM’s learn to reason logically for the task, being able to pick up nuances of questions and giving well-reflected answers that follow causal chains.

One very clear example is arithmatics. LLMs can multiply numbers correctly if the numbers are small. When the numbers get bigger, the result is often wrong, but not off by much, rather a good guess of the actual answer. The threshold for correct arithmetics gets higher with more compute and training data.


The core skills that make me an employee of exceptional ability are:

  • high problem-solving ability,
  • quick adaption to new knowledge,
  • strong ownership of projects/high agency,
  • creative, out-of-the box thinking,
  • extroverted, empathic personality, good communicator.

Illustrative Projects

Here are some example projects that illustarte the aforementioned skills.

Proved a novel Tensor Network Convergence Theorem

My master project was in quantum information theory, and I proved a convergence theorem for a class of tensor network renormalisation schemes.I was especially proud that it happened after my Professor and my Postdoc supervisor had given up on the problem. Solving it involved coming up with a new eigenvector perturbation bound for quantum channels via Perron-Frobenius theory, a field where I had no prior experience in. When I felt like some technique could work, but I lacked the necessary in-depth knowledge, I would contact experts in the field, schedule a meeting with them and asked them questions until I was able to understand the topic well enough to make use of it. The project resultet in a publication in Physical Review Letters, one of the most eminent physics journals.

Reverse-engineered a car security protocol

In 2017, I worked in a car sharing company that developed an interface to remotely open and start the car. My task was to reverse engineer a common (VW) cryptographic scheme that the RFID chip of the car key used. I was able to extract the secret hash function, which allowed us to emulate the protocol on our own hardware for a wide range of cars in circulation. This involved:

  • soldering my own antennas to listen in on the car-key communication,
  • consulting online hacker forums,
  • writing the C code to emulate a car RFID chip with a Proxmark3,
  • finding a security hole that allowed me to probabilistically invert a cryptographic function,
  • acquiring access to an older car (from 1992) with a weaker key length,
  • turning a large truth table into a small Boolean function in algebraic normal form. I was a fresh beginner in all of these tasks and was able to break the encryption within a 4 months time frame. I worked completely independently, but had help with the soldering equipment.

Finished PhD in Mathematics

Doing a master in theoretical physics prepares you only partially for succeeding at writing abstract math papers. Math proofs are much more formal in their description and reasoning. But my colleagues and I fromed a grat team and largely supervised and supported each other. We have published papers in widely varying areas of computational complexity theory, probability theory, theory of metric spaces and applied machine learning at top journal and conferences such as Nature, ICML and AISTATS. I was especially proud of coming up with my own research agenda and developing novel proof techniques in fields that were new to me, often binge-watching university lectures about obscure topics late at night. Over the course of my PhD, while working at Charité, Technical University Berlin or the Zuse Institute, I supervised 7 bachelor, master and PhD students and help them succeed as well.

Presentations and Outreach

Throughout my studies, I worked as an educator at the university for . I gave talks at conferences including ICML, AISTATS, XAI World, Sampta and DPG. I was furthermore invited to talk in seminaries of AIMS (Capetown), GAMM, Mathematics of Deep Learning. In 2020, I performed in the TU Science Slam as part of the Long Night of Science.

I have on numerous occasions received the feedback that I give highly engaging talks that manage to present complex topics in an intuitive and understandable way. Fitting my presentation to the target audience,

Musical translations

My biggest hobby is musical theatre, and I like translating musical numbers into German. This is a surprisingly intricate and complex optimisation task, as you have to fit meaning, rhythm, rhyming and vocalisation together, which often requires out-of-the-box thinking to make the song sound genuine and not just translated. One of the musicals I translated, I also performed with friends. I am also currently writing my own musical about a startup that creates a rogue AI, more infos on my website.

Puzzle hunts

I love solving and designing and logical puzzles. I participated in multiple puzzle hunts (2x Amazon Internal Puzzle hunt, was invited by friends who work there, 1x MUMS puzzle hunt, countless times informal ones with friends). Four time in the last 10 years, I designed or co-designed puzzle hunts or murder mystery games that I then played with my friends or colleagues.