My education, research and industry experience — from electrical engineering at Paderborn University to Ph.D. research at TU Dortmund and a research sprint at the SETI Institute.
All languages that I speak, or am currently learning.
German
Mother language
English
Highly proficient
Chinese
A1 · Duolingo score 32
Profile
I build multimodal foundation models — from custom tokenizers to billion-token training runs to peer-reviewed results.
Machine-learning researcher and engineer specialising in multimodal transformers and computer vision for complex geospatial data. I own the full pipeline: data engineering for millions of image patches, custom VQ tokenizers, distributed PyTorch training on multi-GPU hardware (1.3B+ tokens), and evaluation rigorous enough to hold up in peer review — with experience taking models from research code to robust training and deployment workflows on Google Cloud. I will complete my Ph.D. at TU Dortmund by the end of 2026 and am actively looking for machine-learning roles.
Timeline
Experience
Education & work experience, most recent first.
ResearchPresentPh.D. expected: end of 202606/2022 – presentDortmund, Germany
Researcher & Ph.D. Student
TU Dortmund University · Image Analysis Group
Built a 29.7M-parameter any-to-any multimodal transformer (masked autoencoder) over grayscale imagery, DEMs, surface normals and albedo maps — trained on 1.7M+ image patches / 1.34B tokens; predicts DEMs at 0.945 SSIM and 52.17 dB PSNR (0.077% relative error). First-author Q1 journal paper, ISPRS J. Photogramm. Remote Sens. 2026.
Architected a multimodal masked autoencoder fusing 12 remote-sensing modalities across 56 lunar swirl sites — 191k co-registered patches, 48M tokens, custom VQ-VAE tokenizers at >0.96 SSIM — probed via Leave-One-Out ablations and representational similarity analysis (under review, Computers & Geosciences).
Benchmarked three state-of-the-art anomaly-detection architectures on 1M+ LRO NAC image patches: 98.2% accuracy / 98.7% AUC at the Apollo 15 site and raised the prior baseline's average precision from 49.0% to 62.3% at Apollo 17 (VISAPP 2025).
Run end-to-end training pipelines in PyTorch (DDP, bf16 mixed precision, torch.compile) on dual NVIDIA A6000 GPUs; 4 first-author publications during the Ph.D. so far.
Research06/2024 – 08/2024Mountain View, CA, USA
FDL Researcher
SETI Institute · Frontier Development Lab
Selected for the NASA-backed Frontier Development Lab — an intensive 8-week AI research sprint in Silicon Valley with an international team of researchers and industry partners.
Applied machine learning to global lunar hydration mapping; results presented in two AGU 2024 contributions and an LPSC 2025 abstract.
Built automated defect detection and classification pipelines for natural wood surfaces (SPION project) — from classical image-processing feature extraction to supervised and semi-supervised deep learning.
Published two first-author IEEE conference papers while still studying (ICIT 2021, SSI 2022); the work led to a follow-up ICIT 2024 paper and a Springer book chapter.
Education04/2019 – 04/2022Paderborn, Germany
M.Sc. Electrical Engineering
Paderborn University
Specialised in computer vision and machine learning. Thesis: anomaly detection on natural wooden surfaces.
Education10/2015 – 03/2019Paderborn, Germany
B.Sc. Electrical Engineering
Paderborn University
Foundation in digital systems, signal processing and embedded systems. Thesis: characterization of PCB magnetic fields.
EducationWork08/2013 – 08/2015Soest, Germany
Specialised A-Levels & Internship
Börde Berufskolleg & Legrand
Technically focused A-level qualification, alongside a year-long internship at Legrand — first hands-on experience in electrical engineering and industrial processes.
Off the clock
Hobbies
What I do besides working.
Bouldering
Physical and mental problem-solving: strength, flexibility, and the satisfaction of cracking a new route.
Wakeboarding
Summer on the water: learning new tricks, hitting kickers, and pushing limits one session at a time.
Reading
Sci-fi and popular science in English: both entertaining and a great way to keep growing intellectually.
Electronics
Tinkering with circuits, microcontrollers and DIY builds, where engineering curiosity meets hands-on making.
German nationality
