About
I build machine learning systems that turn physiological signals and medical images into clinical intelligence.
As a Postdoc at LMU University Hospital Munich, my research centers on machine learning for clinical signals and images: representation learning on ECG, wearables, and coronary angiography; LLM systems for structured extraction and quality appraisal in clinical research; benchmarking, validation, and explainability of time-series models; and privacy-preserving computation across institutions.
Dr. rer. nat. (summa cum laude); doctoral research in computational neuroscience. Interdisciplinary background in engineering, medicine, and sport science.
Research Focus
Perceive
Representation learning on clinical signals and images: self-supervised video models on coronary angiography, ECG foundation model adaptation, and multimodal wearable biosignals (ECG, PPG, EDA, accelerometry) for seizure detection and prediction
Trustworthy
Systems whose certainty and reasoning can be checked: benchmarking and validation against clinical references, explainability for time-series models, systematic evidence appraisal, and privacy-preserving computation across institutions (SMPC, federated computing)
Communicate (in progress)
LLMs as a communication and reasoning layer over perception models, literature, and clinical tools — following a propose–verify–approve pattern already piloted in LLM-assisted evidence workflows
Open Source Projects
A selection of recent work — more on GitHub
secure-median
A Secure Median Implementation for the Federated Secure Computing Architecture — companion code to Applied Sciences publication
PaperMole
Open-source, multi-step LLM pipeline for structured extraction and quality appraisal from scientific literature — schema validation, self-verification, human-in-the-loop review (PaperMole). Powered a 196-study systematic review with quantified quality control.
pi-mpc-demo
Secure Multi-Party Computation implementation on Raspberry Pi hardware for privacy-preserving distributed computing
claude-code-docker
Secure, isolated Docker environment for Claude Code with complete filesystem isolation and non-root execution
ICTALS2025
Workshop materials for the International Conference on Technology-Assisted Learning of Seizures — wearable-based seizure detection
pydmdeeg
Dynamic Mode Decomposition for EEG signal analysis — extract spatio-temporal coherent patterns from neural recordings. Validated in 4 peer-reviewed publications.
pet-clinic
Learning repository exploring Privacy Enhancing Technologies (PETs) including SMPC, federated learning, and differential privacy
Publication Code
Experience
Postdoctoral Researcher — AI-Based Telemonitoring
LMU University Hospital Munich
Visiting Scientist — DAAD Fellow
Universidad Francisco de Vitoria & Hospital Universitario La Paz, Madrid
Research Associate
Paderborn University, Institute of Sports Medicine
Student Research Assistant
Paderborn University, Institute of Sports Medicine
Education
Selected Publications
-
Toward clinical-grade deep learning in coronary angiography: a systematic review
Under review at npj Digital Medicine
-
DOI
Challenges in the application of ECG foundation models to out-of-domain populations: an evaluation in professional athletes
EHJ – Digital Health, 7(Suppl. 1), ztaf143.017
-
DOI
Privacy-friendly evaluation of patient data with secure multiparty computation in a European pilot study
npj Digital Medicine, 7(1), 280
-
DOI
A Secure Median Implementation for the Federated Secure Computing Architecture
Appl. Sci., 14(17), 7891
-
DOI
Classification of age groups and task conditions provides additional evidence for differences in electrophysiological correlates of inhibitory control across the lifespan
Brain Inform, 10(1), 11
-
DOI
Classification characteristics of fine motor experts based on electroencephalographic and force tracking data
Brain Res, 1792, 148001
-
DOI
Classification of visuomotor tasks based on electroencephalographic data depends on age-related differences in brain activity patterns
Neural Netw, 142
Selected Talks
- 2025 Challenges in the application of ECG foundation models to out-of-domain populations ESC Digital Health / AI Summit, Berlin
- 2025 Collaboration without data sharing: the Federated Secure Computing architecture deRSE25, Karlsruhe
- 2024 Machine learning approaches to study signatures of age-related brain alterations Colloquium, Universidad Francisco de Vitoria, Madrid
- 2024 Decoding the aging brain Seminar Series AI and Data Science, LMU Munich