Machine Learning Engineer · Munich, Germany
Building AI that works
in the real world.
I build and evaluate machine learning systems at Linde and study Data Science at LMU Munich. My work connects reliable AI research with practical applications in industry and healthcare.
Machine Learning Engineer @ Linde/M.Sc. Data Science @ LMU Munich
From careful evaluation
to useful systems.
When do foundation models beat classical forecasting? I evaluated six models against statistical baselines on 1,118 industrial time series, using rolling-origin evaluation to understand where each approach works best.
Built a pipeline that turns documents, slides, spreadsheets, and images into structured Markdown. Alongside this, I evaluate retrieval, reranking, OCR, and multilingual embeddings to investigate enterprise RAG quality and failure modes.
Investigated whether black-box membership-inference attacks can reveal training-data membership in synthetic clinical language models, including how privacy risks vary across patient subgroups and data conditions.
Combined pose estimation, object detection, and sequence modeling to detect swing phases and generate biomechanical feedback. Deployed a Dockerized FastAPI service on Google Cloud Run and integrated it with an iOS app.
Research-minded.
Built for production.
I work on industrial AI evaluation, document understanding, and satellite object detection. Replacing a YOLOv8 baseline with RF-DETR Large improved detection from 84.5% to 92.0% mAP@50.
My work also covers reproducible training and evaluation with Airflow and MLflow, plus optimized inference with ONNX, TensorRT, and Triton.
As the sole ML engineer at healthcare startup MknoonAI, I owned the workflow from dataset curation to deployment. Models included chest X-ray classification (95.5% F1) and dual-view mammography (85% F1), with tumor-localization support.
Achieved Upwork Top Rated status with a 100% job success rate.
Understanding models,
including their limits.
IEEE Access · 2024
We studied the adversarial robustness of concept bottleneck models, comparing them with conventional CNNs and examining the effects of training strategy and conceptual complexity.
Read the paperContributed log1p support to the PyTorch frontend of the Ivy open-source library through a merged pull request.
Curiosity, with a
practical purpose.
I’m a machine learning engineer based in Munich, with a background in computer science and a focus on reliable AI. I enjoy working across the full ML lifecycle: understanding the data, asking how a model fails, and turning promising ideas into useful, reproducible systems.
My experience spans medical imaging, industrial forecasting, computer vision, and document intelligence. That mix shapes my interest in the same question: how can we make AI systems more dependable in the settings where people actually use them?
LMU Munich · Elite Graduate Program
Konrad Zuse School of Excellence in Reliable AI Scholar
Innopolis University
GPA 4.8 / 5.0
First place, InnoGlobalHack at Innopolis University. More on GitHub
Python · PyTorch · Transformers · OpenCV · scikit-learn · Docker · FastAPI · Airflow · MLflow · ONNX · TensorRT · Triton · GCP · AWS
Let’s connect
Interested in reliable AI, research, or building something useful?
I’d be happy to hear from you.