I started in molecular biotechnology and I'm now working through an M.Sc. in Bioinformatics at DTU Copenhagen. In between, I build things for the web — often where biology and interfaces meet.
Lately I've been reading into synthetic proteins — designing sequences instead of just classifying them. My thesis compared Transformers and xLSTMs on protein data for function prediction, and it left me curious about the opposite problem: predicting what a sequence does is one thing, writing a new one that actually folds and works is a much harder bet. No real project yet, just a growing stack of papers on things like RFdiffusion and ProteinMPNN, and a hunch this is where I want to push next.
In progress. Moving from wet-lab molecular biology into the computational side — sequence analysis, statistics, and machine learning applied to biological data.
Built a Unix-based HPC pipeline for analyzing 2nd- and 3rd-generation sequencing data, under Michael Westbury.
"Gene Ontology Term Prediction from Protein Sequences: Transformers vs. xLSTMs." Trained both architectures on HPC infrastructure under Dr. Pahle, comparing how well each predicts protein function from sequence alone.
Foundational degree in molecular biology, biochemistry, and biotechnology — the wet-lab base everything above builds on.
Commissioned build combining Java, HTML, and Python to visualize genomic data for the client.
Statistical analysis of lipidomics data, extended internal R packages, and built Salesforce API integrations, under Dr. Mathias Gerl.
A student redesign of NCBI's BLAST, built for a course called COMP4020. The original's dense tables and tiny type hadn't kept pace with the rest of the web — this keeps the same workflow (paste a sequence, pick a search type, read ranked hits) with a much clearer interface.
An interactive teaching tool for the HP (hydrophobic–polar) model of protein folding, built for COMP4020. Fold a simplified amino-acid chain by hand with the arrow keys, watch the energy score change with every move, then compare your result against the computed optimal fold to see how closely you matched water's preferred configuration.
A browser instrument built for COMP4020, rendered with Three.js over the raw Web Audio API. There is no score and no note grid — you drag lily pads, mushrooms, rocks and stumps onto a small pond, and where each one lands is what decides how it sits in the loop. Seven voices, from soft pads through to kick and snare. Click the windmill to stop the whole thing.
