Projects

Personal project

A web app for residents of a shared kitchen to organise cleaning duties and split shared expenses. I designed a React interface and a Python API in FastAPI with a PostgreSQL database.

GitHub Actions tests each change, builds a Docker image, deploys it to Google Cloud Run and checks database access after each release.

Open the live demo

Master’s thesis, Copenhagen Business School with RAIT, Bispebjerg Hospital

Machine Learning for Radiographic Assessment: Investigating Crossing-Sign Recognition on Lateral Knee Radiographs. I worked with clinicians to understand what makes a feature difficult to identify on knee X-rays, and to agree on what an automated assessment needs to capture to support their judgement.

We defined the training data requirements together, and I built a two-stage pipeline whose results practitioners can interpret. Across the models I evaluated, the best reached a median error of 0.51 mm.

Team project, Advanced Machine Learning for Computer Vision, IT University of Copenhagen

We fine-tuned YOLOv8 models to detect surface defects on circuit boards. Configuration files made the training runs repeatable, and we tracked detection metrics in Weights & Biases.

We compared saved model checkpoints and chose a detection threshold on validation images. We then measured missed defects and false alarms on a separate test set.

Code on GitHub

Team project, Advanced Machine Learning for Natural Language Processing, IT University of Copenhagen

We automated four training runs on the Stanford Natural Language Inference dataset, and logged accuracy, training time and peak GPU memory to compare fine-tuning methods.

DistilBERT reached 83.48% test accuracy with low-rank adaptation (LoRA) and 81.12% with partial fine-tuning.

Code on GitHub