Research Scientist · TUM University HospitalApr 2024 – PresentLLM systems for clinical reporting, information extraction and agentic decision support.
Munich, Germany
Developed a customer-facing medical report generation system with automated billing. The system extracts clinical components and billing rules from the data so that the report content is medically accurate and correctly coded. Worked directly with physicians on requirements, feedback and iteration.
Built LLM/NLP pipelines to extract and structure rare medical information from unstructured clinical text.
Research on observability and verifiability of agentic clinical decision support systems, in industry cooperation with Springer Medicine.
NLP
LLM
Agentic systems
OpenTelemetry
PyTorch
HuggingFace
LangChain
LangGraph
Async Python
OpenMP
OpenACC
CUDA
PostgreSQL
Docker
GCP
AWS
TypeScript
React
REST
FastAPI
RAG
pgvector
Pydantic
Fine-tuning
vLLM
Ollama
CI/CD
Git
Research Scientist · Technical University of MunichDec 2022 – Mar 2024Hybrid physics–machine learning models with differentiable FEM solvers.
Munich, Germany
Developed differentiable finite element (FEM) solvers in JAX and coupled them with graph neural networks and score-based diffusion models to build hybrid physics–machine learning models.
Python
JAX
Flax
Optax
Jraph
SLURM
HPC
PyTorch
PyTorch Geometric
GNNs
FEM
Diffusion models
Working Student · Technical University of MunichSep 2021 – Mar 2022Uncertainty estimation for transformer-based perception in automated driving.
Munich, Germany
Researched uncertainty estimation for transformer-based object detection and panoptic segmentation (DETR) on simulated traffic scenes with CARLA, to improve the reliability of perception in automated driving.
Computer Vision
Uncertainty Quantification
Vision Transformers
CNN backbones
PyTorch
OpenCV
W&B
Software Developer Intern · BoschJun 2019 – Jan 2020Functional software for radar- and camera-based predictive safety (ADAS) systems.
Budapest, Hungary
Developed functional software for radar- and camera-based predictive safety (ADAS) systems.
C
C++
Software-in-the-Loop
Bitbucket
Jenkins CI/CD
DOORS
CAN signals
ISO 26262
Education
PhD · Trustworthy and Efficient Medical Language ModelsApr 2024 – PresentTechnical University of Munich (TUM) · Supervised by Prof. Daniel Rueckert.
Institution
Technical University of Munich (TUM), Munich, Germany
Supervisor
Prof. Dr. Daniel Rueckert
Ongoing projects
Aligning Language Models through Differentiable Self-Discrimination
Quantifying the Language Barrier in LLM Knowledge Integration and Inference
Reliable Traces for Agentic Clinical Evidence from Large Language Models
Building Medical Knowledge Graphs from Text and Tabular Data for Grounded Predictions
Master of Science · Robotics, Cognition, IntelligenceOct 2020 – Nov 2022Technical University of Munich (TUM) · Graduated with distinction (top 14%).
Institution
Technical University of Munich (TUM), Munich, Germany
Grade
Graduated with distinction (top 14%)
Specialization
Deep Learning, Data Science, Generative AI
Thesis
Invariance and Equivariance in Machine Learning for Physics
Supervisor
Dr. Marten Lienen
Master of Science · Mechatronics EngineeringFeb 2020 – Jan 2022Budapest University of Technology and Economics (BME) · Excellent with highest honors (top 5%).
Institution
Budapest University of Technology and Economics (BME), Budapest, Hungary
Grade
Excellent with highest honors (top 5%)
Specialization
Intelligent Embedded Mechatronic Systems
Thesis
Sentiment analysis of text sequences and session-based recommender systems
Supervisor
Prof. Dr. Luca Szegletes
Bachelor of Science · Mechatronics EngineeringSep 2016 – Jan 2020Budapest University of Technology and Economics (BME) · Excellent with highest honors (top 5%).
Institution
Budapest University of Technology and Economics (BME), Budapest, Hungary
Grade
Excellent with highest honors (top 5%)
Thesis
Objectified hazard and risk analysis of predictive safety systems (in cooperation with Bosch)
Exchange semester (Oct 2018 – Feb 2019) at the Karlsruhe Institute of Technology (KIT) with a Campus Mundi Scholarship.
Teaching & Supervision
Courses
WS26Large Language Models in Medicine · Master's Seminar
WS23Advanced Deep Learning for Physics · Exercises and exam
Supervised Master's theses
WS26Building Medical Knowledge Graphs from Text and Tabular Data for Grounded Predictions
WS26Actor–Critic Agent Architectures for Safe Clinical Decision Support: Design and Evaluation on Medication Safety Review
WS26Continuous Learning in Medical Documentation: Agent-Driven Extraction for Automated Operative Reports
WS25Aligning Language Models through Differentiable Self-Discrimination
SS25Predicting Academic Impact through Semantic Text Analysis
WS24–SS25Reducing Unintended Personally Identifiable Information Retention in Fine-Tuning Large Language Models
WS24A Comparative Analysis of Retrieval-Augmented Generation and Fine-Tuning in the Biomedical Domain
Fellowships & Grants
CLARITY: AI-based Contextual Artifact Reduction for Integrity and Traceability in Multimodal Clinical RecordsBMFTR Zukunft eHealth Grant · €2 million · 2027–30 · co-authored
