ML Engineer · AI Researcher · Builder
I recently completed my B.Sc. in Artificial Intelligence and Data Science at OTH Regensburg (March 2026) and will be starting my M.Sc. in Data Science (Machine Learning and Statistics track) at Uppsala University, Sweden in September 2026.
My research focuses on deep learning, computer vision, and Vision Transformers — particularly for document understanding and structured data extraction. I'm also the founder of Casdo Labs, where I'm building Patronus, an AI security platform for LLM monitoring, red-teaming, and safety guardrails.
- 🎓 B.Sc. AI & Data Science — OTH Regensburg (Completed March 2026)
- 🎓 M.Sc. Data Science (ML & Statistics) — Uppsala University (Starting Sept 2026)
✈️ Mitacs Globalink Research Intern — University of Alberta, Canada (May–July 2026, DAAD Scholarship)- 🔬 Research: Table structure recognition, document digitalization, multimodal AI
- 🏗️ Founder: Casdo Labs — building Patronus (AI security for LLMs)
| Project | Institution | Period |
|---|---|---|
| VisLang-Seg: Vision Transformers & LLMs for multimodal image segmentation | Mitacs · University of Alberta, Canada (DAAD Scholar) | May–July 2026 |
| Bachelor Thesis: ViT-based table structure recognition with geometry-aware loss function | OTH Regensburg | Oct 2025–Mar 2026 |
| Document Digitalization using deep learning & computer vision — Swiss ornithological records; poster at DOG Conference 2025 | OTH Regensburg | Sep 2024–Present |
Vision Transformer-Based Table Structure Recognition with a Novel Geometry-Aware Loss Function
Introduced a geometry-aware loss formulation extending conventional BCE–Dice training with three complementary terms: axis-alignment (orientation-sensitive gradients), morphological closing (connected, hole-free regions), and spur-suppression (differentiable opening operator). Evaluated on PubTables-1M — gains of up to +9 percentage points in region-level F1 over the BCE-Dice baseline, without inference-time overhead.
- trinnovative GmbH — Working Student, Data Science & Backend (Apr 2025–Present)
- OTH Regensburg — Tutor, CodeClinic Programming (Oct 2025–Present)
- Audi AG — Working Student, Data Science & AI (Oct 2024–Mar 2025)
- Data mining for error sequence analysis; unsupervised driving style recognition (autoencoders + clustering); LLM assistants with document knowledge bases
- AVL Software and Functions GmbH — Working Student, AI Engineering (May–Aug 2024)
- LLM-powered autocompletion with N-gram/Trie methods; international team collaboration
- Siemens AG — Mechatronics Apprentice (Sep 2018–Aug 2021)
Patronus — AI Security Platform (Casdo Labs)
Enterprise LLM monitoring, automated red-teaming, and safety guardrails for production AI systems.
Document Digitalization using Deep Learning & Computer Vision
Extracting structured data from handwritten historical tables. Digitizing Swiss ornithological records for research and conservation. Poster presented at DOG Conference 2025.
Schula — AI Feedback for Student Essays
Educational LLM tool that compares student essays with learning goals and provides formative feedback for self-assessment.
Driving Style Recognition (Audi AG)
Unsupervised deep learning model to analyze driving behavior from time-series vehicle sensor data.
Languages
ML / AI
Data & Infra
Always open to collaborations in machine learning, computer vision, and AI research. Let's connect!