About me
I'm a data scientist who has worked in healthcare and the life sciences since 2017, turning messy, multi-source data into decisions people can use. The question I keep coming back to is a simple one: can a clinician rely on this model for a patient it has never seen?
Background
- BEng in Bioinformatics and an MSc in Data Science at the Silesian University of Technology in Poland, both graduated with distinction.
- PhD in Bioinformatics at the Technical University of Munich, on epigenomics and machine learning. I submitted the thesis in April 2026.
- Now a postdoctoral data scientist at LMU University Hospital in Munich, working on AI for oncology and precision medicine.
- Experience across research and industry since 2017, including a software engineering role building hospital SQL reporting and several open-source scientific tools.
Currently
At LMU University Hospital I build machine learning for oncology. That covers computational-pathology models that read whole-slide images, survival and decision-support tools, and containerized, reproducible pipelines on HPC, working closely with clinicians and national consortia (DKTK, BZKF).
What I work on
- Computational pathology & imaging · Foundation-model embeddings and attention-based multiple-instance learning that predict molecular targets and outcomes straight from H&E whole-slide images and radiology.
- Survival & decision support · Cox-regression tools and risk models a clinician can read at the bedside, built to turn high-dimensional data into treatment planning.
- Multi-omics & epigenomics · Methylation and multi-omic integration, with open-source tools (jDMRgrid, DMRspiker) and benchmarks that make differential-methylation results reproducible.
- Reproducible ML / MLOps · Containerized pipelines (Docker and Apptainer) on HPC, external validation done properly, and evaluation that holds up on a new cohort or scanner.
A bit more
I work in Polish, English, and German. I like problems where the whole range matters: engineering, modeling, and explaining results to people who do not build models.
Prefer the short version? Download my CV.