Austria
I am a research mathematician with core interests in theoretical and applied machine learning, deep learning, mathematical data analytics, formal methods, and artificial intelligence. In 2024, I was awarded a doctoral degree (DPhil/PhD) from the Mathematical Institute of the University of Oxford, where I also held a subsequent postdoctoral research position as recipient of Oxford's InFoMM Doctoral Prize Scheme. I am currently working as a Senior Research Scientist for Beneficial AI Foundation, reporting to Prof. Max Tegmark (MIT). In addition, I offer independent research consulting on technical AI topics and high-level AI strategies. I am experienced in designing novel neural architectures, conducting systematic computational experiments, proving theorems in theoretical deep learning, formally verifying software libaries via Lean 4, modelling biological and chemical phenomena, conducting research in drug discovery, and working with a wide range of data types, including graphs, time series, and SDE-simulated data. I have multi-year experience collaborating with industrial partners, and communicating effectively with both experts and non-experts alike. My programming expertise spans Python, PyTorch (with GPU acceleration), Lean 4 for formal verification and interactive theorem proving, and many other frameworks. I have a publication record of technical and scientific contributions across a wide set of scientific AI domains. Selected past projects include: ▪ Using LLM-based tools and Lean 4 to formally verify elliptic-curve cryptography software libraries (currently). ▪ Developing a new deep learning solver for high-dimensional PDEs without curse of dimensionality (NeurIPS 2020). ▪ Designing original math problems to investigate the reasoning limits of LLMs (Epoch AI FrontierMath 2024). ▪ Benchmarking graph isomorphism networks vs. classical mol. featurisers for activity cliff prediction (JChem 2023). ▪ Predicting the onset of chaotic dynamics in time series data using neural networks (collab. with CCFE 2020). ▪ Investigating high-impact research frontiers in deep learning for video games (Beam Foundation 2026). ▪ Implementing an improved version of a classical substructure-based molecular feature-extractor (JChem 2024). ▪ Designing a novel graph-based twin neural network architecture for activity cliff prediction (AIChem 2021). ▪ Using horizontal visibility graphs to detect neonatal brain injury in biomedical time series data (SMNS 2017). ▪ Employing transfer entropy for causal inference in complex metabolomic networks (at UCL CoMPLEX 2017).
Using Lean 4 and novel LLM-based tools to investigate formal verification in AI safety and the automation of mathematical proof.
Advising companies, start-ups, non-profit organisations, and individual professionals on technical AI solutions and AI strategies.
Designing original, difficult math problems for FrontierMath benchmark to test reasoning limits of LLMs
The aim of InFoMM's Doctoral Prize scheme is to help Oxford retain the best doctoral students to increase the academic impact of their doctorates and to assist in launching their research careers. This is achieved by enabling recently graduated doctoral students who are of the highest academic calibre to undertake short projects (up to six months) as postdoctoral research associates at the Mathematical Institute.
Collaborating with my partner company Lhasa Limited as part of my doctoral research project at the Mathematical Institute of the University of Oxford.