I'm a computer scientist passionate about building thoughtful, well-crafted software at the intersection of web development, machine learning, and data visualization. I enjoy turning complex ideas into clear, usable systems: whether that's an interactive interface, a model that explains itself, or a visualization that makes patterns visible at a glance.
Currently, I research uncertainty quantification in large language models, studying how these models and methods work while benchmarking different approaches. My focus is on understanding, evaluating, and improving the reliability and interpretability of machine learning systems.
Previously, I studied computer science alongside teaching and education, which shaped how I approach both research and development. I've worked on a range of projects spanning web applications, ML pipelines, and educational tools, always with a strong emphasis on clarity, accessibility, and knowledge sharing.
Outside of work, I’m usually hiking through forests or mountains, building things by hand, experimenting with code and visualizations, or writing poetry. I'm queer, naturally curious, and happiest when learning something new or creating something tangible.
Helmholtz-Zentrum Dresden-Rossendorf
Researching uncertainty quantification in large language models. Prepared and taught an LLM workshop at the HZDR HPC Fall School 2025.
Technische Universität Dresden
Master’s thesis: “Uncertainty Estimation of Large Language Model Replies in Natural Sciences”
Technische Universität Dresden
Specialization: Upper secondary education (Mathematics and Computer Science)
Westsächsische Hochschule Zwickau
Bachelor’s thesis: “Comparison of Angular and Vue.js for introducing component-based frontend frameworks in an academic course”
This poster presents a modular benchmarking framework designed to ensure reproducibility and transparency when evaluating probabilistic and resource-intensive AI systems, with a focus on uncertainty estimation for large language models. The framework decouples model execution from evaluation logic, enabling consistent, reusable, and longitudinal benchmarking as models and APIs evolve.
This poster presents a qualitative and quantitative evaluation of uncertainty metrics for large language models in scientific question answering. It assesses a range of theoretical approaches and benchmarks a subset of metrics for reliability, highlighting the strengths and limitations of token-level, verbalized, and consistency-based measures. The poster was awarded second place for Best Poster in an audience vote among 200 entries.