As a freelancer I worked together with BCG GAMMA on a carbon accounting app CO2AI. I created an emission factors browser, a system for managing API access tokens, and an object storage middle layer. Choosing the right emission factors is a major problem in carbon accounting. So, the emission factor browser uses NLP and good UX to make finding the right emission factor efficient. By now the team doubled-down on this app and carved out a separate company for it (co2ai.com).

As part of a DevOps team I created a serverless, GxP-validated, pharmacometrics IT landscape. We had to create fully validated and monitored computation environments for different statistical languages. All data transformations statisticians would perform, had to be tracked and stored in a browsable graph. Additionally, I mentored colleagues in software development and helped recruitment efforts with tests and interviews.

This is a side-project I did on the weekend. It's an interactive pilot chart for surfers, pilots, sailors, paragliders, and anyone else interested in wind strength and directions. There are many great apps, like windy.com, that show you the current winds with short-term predictions. However, I didn't find anything with historic wind data for planning a holiday or passage.

In a machine learning group focused on drug discovery I was developing a biologics transformer model. It was based on BERT, a model commonly used for natural language processing tasks. I re-purposed it to understand amino acid sequences of antibodies. I designed and performed pre-trainings with about 1 billion amino acid sequences and further follow-up fine-tunings on different downstream tasks. The transformer was on a par with state-of-the-art antibody transformers around that time.

This was a little side-project I did to help beekeepers. I developed and trained a model which can detect the content of a honeycomb photo. E.g. it can count the number of cells with larvae, pollen, nectar and so on. You can use this to get an accurate estimate of your beehive's health.

In a DevOps team I was developing software solutions for researchers at Bayer, specifically bioinformaticians. My colleagues and I have built a data management system, several visual-analytics platforms, a data-integration, and an algorithms deployment platform. Additionally, we helped bioinformaticians deploy their models. Development, infrastructure, security, and project management roles were shared. Besides that our team was heavily involved in the digital transformation of the organization.

I co-founded and supported Biotop Community Lab e.V. as vice chairman. This is a no-profit organization for promoting the democratization of science through access to biotechnology and learn-by-doing education. We are a part of the global DIY-bio movement, which focuses on bringing biology outside of academic and industrial environment to the lay public. We believe that biology is technology and we want to put citizens in the conditions to make use of it, as any other common technology.

Data from single cell RNA sequencing experiments are often confounded with technical variability due to experimental batches. Using mathematical modeling and machine learning methods I measured these batch effects and predicted significantly differential distributions.

This is a webapp I wrote during my Masters when working with pooled CRISPR/Cas9 screens. Analyzing the data from these screens is tedious and largely repetitive. This tool does all the pre-processing, data quality checks, and differential distribution calculations. Years later I noticed this tool was used by several Pharma companies.

I worked in a microfluidics laboratory which wanted to create synthetic cells using micro-droplet technology. For this surfactant mixtures had to be optimized. I helped establishing a system which could measure surface tension and other physical properties directly inside the micro-droplets using computer vision.

A Singaporean Start-up found a set of genes which were differentially expressed in leukocytes of sepis patients. I facilitated a proof of value study by analyzing patient blood samples on call. The work included leukocyte mRNA extraction and qPCR.

This thesis combined spatial simulations and synthetic biology to compare static and spatio-temporal oscillating concentration gradients within the cell. The work included modeling experiments in silico and then performing them in vivo.