Agentic generation of web-backend security benchmark tasks with functional tests, contrastive exploits, and an executable benchmark release.
Research / Writing / Projects
Research
AutoBaxBuilder: Bootstrapping Code Security Benchmarking
ICML 2026
Memorization Dynamics of Fill-in-the-Middle Pretraining
MemFM @ ICML 2026
Controlled 100B+ token pretraining experiments, studying FIM vs. left-to-right memorization through probabilistic extraction and attention patterns.
Neural Networks for Signal Deconvolution in Cancer Transcriptomics Data
BSc thesis at Boeva Lab, ETH Zurich, 2023
Developed novel autoencoder and variational autoencoder approaches for cell-type deconvolution from bulk RNA-seq data.
Writing and notes
- On being human in a world of machines 29 Jul 2026
- AlgoLab Fall 2024, ETH Zürich Algorithms Lab, C++
- Randomisierte Algorithmen 15 Aug 2021, German
- Analysis Cheatsheet 27 Jul 2021
- Intuitive Approach to Solve Probability-DP Tasks 3 Jul 2021
Projects
crosscheck
2026, Juno Hackathon
Crosscheck tests whether health AI gives consistent advice when the same symptoms are described differently, and which facts change that advice.
HTML5-JS
2017–2020, HTML5 and JavaScript
A collection of small browser experiments: fractals, games, graphics, an audio sampler, and a neuroevolution racing simulation.