Computer Vision · VLMs · 3D Reconstruction · Medical AI
I am an M.Sc. computer science student at TUM and a computer vision research working student at ImFusion. My work there spans surgical 3D reconstruction, visual odometry, agentic medical imaging, and applied machine learning.
Currently
Developing real-time 3D reconstruction and SLAM for monocular surgical endoscopy in Munich.
Experience
May 2026 - present
ImFusion, Munich
Developing real-time 3D reconstruction and SLAM methods for surgical endoscopy from monocular video. I also built a pip-installable benchmark for quantitative and qualitative evaluation of visual odometry methods on endoscopic data.
Oct 2024 - Sep 2025
Bosch Engineering, Holzkirchen
Continued ML and AI innovation work for steer-by-wire applications, building on my preceding six-month internship in the same team.
Mar 2024 - Sep 2024
Bosch Engineering, Holzkirchen
Developed anomaly detection for water ingress in steering control units, improved the ML pipeline, evaluated time-series and foundation models, and built a GUI for expert validation of risk models.
The anomaly detection algorithm is now patented (2025/1349)
Jan 2022 - Aug 2022
TRIORAIL Bahnfunk, Pfaffenhofen
Developed and maintained an Android app used by engineers servicing railway communication stations, including a PDF export workflow.
Selected projects
Recent projects at the intersection of machine learning, medical data, computational biology, and software engineering.
TUM, Machine Learning in Medical Imaging · 2026
Built an agentic VLM system that reasons over whole-slide pathology images and generates reports. The graph-based workflow uses ReAct-style revision, evidence tools, CONCH retrieval, HybridRAG memory, and a LoRA-tuned Qwen VLM.
60% final diagnosis accuracy and a 12% gain from fine-tuning
View project →
Claimini GmbH, TUM Data Innovation Lab · 2026
Created a two-stage classification and regression model for repair-cost prediction from about 100,000 real insurance claims. The work included extensive preprocessing, calibration, and analysis of high-cost drivers and repair-shop allocation.
Estimated €2.8M in resources freed per year
View project →
TUM seminar, Vision-Language Models in Medical Imaging · 2025
Benchmarked Qwen2.5-VL-72B-Instruct on chest X-rays and brain MRIs across classification, visual grounding, description, detection, and diagnosis, with prompt designs tuned for consistent structured output.
View project →
Gagneur Lab, TUM · 2025
Developed a BPNet-style multitask CNN for simultaneous RNA expression-profile prediction. The dual-head model combines pretrained SpeciesLM embeddings with total transcript-count and base-resolution shape prediction.
View project →
Rost Lab, TUM · 2025
Developed a ProtENN2-style Pfam classifier and compared ProtT5 residue embeddings with one-hot inputs. I assembled the dataset from Pfam, handled class imbalance, and analyzed domain-level clustering.
University of Pisa, Computational Mathematics for Learning and Data Analysis · 2026
Implemented a primal-dual Interior Point Method for constrained Lasso least squares from the ground up, with a full KKT system solver and a reduced variant based on the Schur complement. This is the numerical optimization groundwork that every deep learning method rests on.
Bachelor's thesis, TUM · 2023
Designed and executed a simulation-based benchmark suite for modular robots and added robotic pick-and-place features to the open-source Timor Python toolbox.
Thesis grade 1.3
View project →
hackaTUM, CHECK24 challenge · 2023
Built a full-stack MVP for geo-distance-based craftsmen ranking in a team of four. We optimized both the weighted ranking algorithm and database access, then presented the result at CHECK24.
View project →
About me
I study computer science at the Technical University of Munich, specializing in digital biology and medicine, machine learning and data analytics, and software-intensive systems. In 2025/26, I spent an Erasmus+ semester at the University of Pisa working on digital health AI and computational mathematics for learning (interior point methods, KKT systems, Schur complements, and Lasso least squares).
I enjoy turning research ideas into reliable software and measurable experiments. Outside work and university, I am interested in soccer, volleyball, and skateboarding.
Vision-language models (see my medical VLM evaluation), 3D Gaussian splatting, 3D world modeling, and foundation models for vision and medicine.
PyTorch, TensorFlow, scikit-learn, deep learning, foundation models, vision-language models, LoRA fine-tuning, agentic pipelines, time series, explainable AI
Computer vision, SLAM, visual odometry, 3D reconstruction, 3D Gaussian splatting, 3D world modeling, DICOM, digital pathology, genomics
Numerical optimization (interior point methods, KKT systems), statistics, Python, C++, Java, Rust, SQL, Svelte, Docker, Linux, Git, CI/CD
German (native), English (C2), French, and Italian