Thesis: Towards Closed-Loop Deep Brain Stimulation: A Deep Learning Approach for Real-Time Gait Classification in Parkinson's Disease

In my thesis, I explore deep learning methods for real-time classification of gait states in Parkinson’s Disease using multimodal biosignals (LFP, EEG, EMG, IMU). The goal is to extract robust features and develop predictive biomarkers for integration into adaptive deep brain stimulation (DBS) systems, enabling patient-specific, frequency-based stimulation to improve clinical outcomes.

Relevant Coursework

Thesis: Markerless Pose Estimation and Mice Tracking with Deep Learning.

Developed a deep learning framework for markerless pose estimation and tracking of multiple, interacting mice in long-term behavioral studies. The system integrated RFID sensor data to automatically correct identity swaps, ensuring robust tracking over 24-hour periods and achieving human-level accuracy.

Developing & Benchmarking Inverse Methods for M/EEG Source Imaging

I designed and implemented a modular Python framework to systematically evaluate the uncertainty and calibration performance of brain source reconstruction algorithms. A core aspect of my role involves automating the entire experimental pipeline—from data simulation and forward modeling to evaluation and visualization. This work deepened my expertise in probabilistic modeling, signal processing, and scientific software engineering.

Implementing Methods for Electrophysiological Data Analysis and Signal Processing

My work involved contributing new multivariate brain connectivity methods to an open-source toolbox. This experience deepened my understanding of the principles of collaborative, open-source software development and advanced concepts in linear algebra and calculus.

Developing Toolboxes for Behavioral Data Analysis

My role involved implementing and maintaining R and Python toolboxes for processing and analyzing large-scale behavioral data from RFID-tagged mice in automated home-cage environments. I created statistical analysis pipelines and data visualization tools to support research in behavioral neuroscience.

Hardware Development for eHealth & Agent Systems

My work focused on the design and prototyping of electrical circuits for distributed eHealth systems.

Technical Support & System Administration

Managed the institute's computer labs, providing technical support to staff and students. My responsibilities included hardware and software installation, troubleshooting, and system maintenance to ensure the smooth operation of all IT infrastructure.

CaliBrain: A Python Toolbox for Evaluating Uncertainty and Calibration in M/EEG Source Imaging

As lead developer, I designed and built CaliBrain, a Python toolbox for evaluating uncertainty and calibration in M/EEG source imaging. This library provides a framework for systematically benchmarking inverse solvers to promote well-calibrated source reconstruction methods.

GaitMod: A Python Toolbox for Gait Modulation Analysis

Developed as part of my Master's thesis, `GaitMod` is a Python package designed to analyze multimodal biosignals (LFP, EEG, EMG, IMU) using deep learning algorithms (e.g. LSTM) . The goal is to detect gait modulation in Parkinson's disease and identify biomarkers for adaptive deep brain stimulation (aDBS) for more personalized clinical applications.

ColonyRack: An R Toolbox for Mouse Colony Data Analysis

Developed a framework in R to analyze and visualize complex behavioral data from an automated, large-scale mouse colony system. The toolbox processes continuous RFID tracking data to map social dynamics and movement patterns.

DeepMouseTrack: A Python Framework for Mouse Tracking & Pose Estimation

Developed a Python framework for mouse pose estimation and tracking in a maze environment. This project implements an end-to-end solution using Deep Neural Networks (DNN).

RFID-Scale: A Python Framework for Mouse Weight Estimation

Developed an analysis pipeline to process and visualize weight data from an RFID-based scale system. The pipeline automates estimation of individual mouse weights, and generation of analytical plots.

GroupMouseTrack: A Python Toolbox for Multiple Mouse Tracking

Developed a Python toolbox for markerless pose estimation, tracking, and identification of multiple, identical-looking mice. It provides an end-to-end deep learning solution, leveraging DeepLabCut with a ResNet-50 deep neural network (DNN) in TensorFlow. The system integrates with RFID sensors to automatically correct identity swaps, enabling robust, long-term behavioral analysis without manual intervention.

Contributed to MNE-Python, a leading open-source toolbox for neurophysiological data analysis with over 4,000 scientific citations. My work focused on implementing advanced multivariate connectivity methods in the `mne-connectivity` module, enhancing the library's capabilities for complex signal processing.

Constructing a lead-field matrix corresponding to the AAL2 atlas and standard 10-20 EEG electrode configuration and contributing to the neurolib toolbox for whole-brain modeling.

Utilized deep learning models (UNet and SegResNet) for MRI image segmentation of brain tumors. Implemented workflows to detect MGMT (a DNA repair enzyme) promoter methylation status from MRI images.

Implemented a Multimodal Variational Autoencoder (MVAE) to explore its ability to learn joint representations from multiple modalities to generate images conditioned on labels and vice versa.

As the founder and lead instructor of MathCodeLab, I develop and deliver courses in Mathematics and Computer Science with a focus on programming for undergraduates. To date, I have taught over 10 courses to more than 50 students, building a supportive learning community of over 240 members.

For more details, see the Ventures page or the full list of courses.

As a teaching assistant for this Master's level seminar, I mentored students on advanced topics at the intersection of machine learning and biomedical data analysis. My role involved guiding their literature review, presentation development, and fostering scientific discussion.

As the founder of MathCodeLab, I develop and deliver courses in Mathematics and Computer Science with a focus on programming for undergraduate students. The platform offers tailored online and in-person instruction to provide accessible, high-quality education in foundational STEM subjects.

To date, I have developed and delivered +10 courses to +50 students, building a supportive community of over +240 members across Germany. For more details, see the Ventures page or Courses.

Founded and operate DrinkRack, a hassle-free beverage service for offices and companies. We provide a fully-stocked beverage rack at no upfront cost to the business, operating on a trust-based system where employees pay only for what they consume. Our service covers all setup and regular restocking.

Co-founded an event planning and rental service specializing in creating unique experiences with a blend of Bohemian and minimalist aesthetics. Services include decor design, floral arrangements, and beverage services for various private events.