United Kingdom
Highly motivated Machine Learning researcher with five years of research experience in Computer Vision and NLP and recent PhD graduate in the group for AI in Medicine led by Prof. Daniel Rueckert at the Technical University of Munich (TUM). My key takeaways after years in ML research are: • Every machine learning project starts by understanding the data • Always question evaluation metrics; measuring the right thing matters more than a slight gain on a wrong one • Only a model that is thoroughly evaluated can be meaningfully improved I am looking for interesting research endeavours to use my solid academic background and hands‑on experience to drive impactful technological advancements.
Our team is enabling organizations to fine‑tune Copilot with their own data for their specific needs.
Researcher working on machine learning solutions to improve healthcare.
I'm working on Machine Learning for medical image analysis, especially for weak- and unsupervised detection of diseases and multimodal learning. During my PhD, I • Authored 16 scientific papers • Supervised five master's theses and two semester projects • Led the seminar “Unsupervised Anomaly Detection in Medical Imaging” for three semesters • Co‑organized the BraTS Inpainting Challenge at MICCAI 2023 • Reviewed for MICCAI (best reviewer honourable mention), MIDL, ECCV, CVPR, TMI, and MedIA
• Tutored the popular lecture "Introduction to Deep Learning" with over 1000 participants • Created lecture slides and coding assignments with automatic evaluation via unit tests in Python
• Contributed to the evaluation of MAIRA‑2 — a VLM for radiology report generation — by extracting metadata from unstructured reports using LLMs • Improved RadFact, an LLM‑based evaluation metric, by increasing its interpretability • Achieved 10x inference speed‑up for RadFact by showing performance equivalence of a smaller, open‑source backbone and optimizing inference speed with vLLM and prefix‑caching
• Developed Machine Learning algorithms for electrographic flow estimation and ECG data analysis • Accompanied clinical pre‑trials for a 64‑lead ECG vest