Germany
I am a PhD candidate in Machine Learning for Quantum Chemistry, with a strong background in mathematics and hands-on experience building generative models, equivariant deep learning architectures, and large-scale ML pipelines for molecular simulation and discovery. My research focuses on applying modern machine learning methods to challenging problems in quantum chemistry, including excited-state modeling, reinforcement learning for molecular optimization, and invariant representations of potential energy surfaces. I have published in Nature Computational Science and The Journal of Physical Chemistry Letters, and have several additional works under review. I have industry research experience from IBM Research Zurich, where I worked on foundational models for chemistry, including fine-tuning large language models for tasks such as reaction prediction, retrosynthesis, and spectral interpretation, as well as exploring multi-GPU training strategies at scale. I bring strong software engineering skills (Python, PyTorch), experience working across academia and industry, and a practical mindset focused on turning theory into robust, usable tools. I am finishing my PhD in September and am actively exploring industry roles such as Research Scientist, Applied Scientist, or Research-oriented ML Engineer, particularly in scientific machine learning, AI for chemistry, or foundation models for science.
Machine Learning for Quantum Chemistry. Doctoral candidate supervised by Julia Westermayr at Leipzig University and co supervised by Klaus Robert Muller at TU Berlin
Machine Learning for Quantum Chemistry. Doctoral candidate supervised by Julia Westermayr at Leipzig University and co supervised by Klaus Robert Muller at TU Berlin
Working on implementing GRPO fine-tuning strategies for LLMs in chemistry in the group of Teodoro Laino
Research stay with Professor Christoph Ortner working on parameterizations of conical intersections in potential energy surfaces
My team worked on a project to generate new stable optoelectric molecules given the excitation energies using generative learning methods. This will hopefully reduce the computational cost of these kind of calculations in the future.