Participated in a competitive university hackathon focused on classifying complex, heavily distorted sensor data using ResNet18 (pre-trained then fine-tuned). Engineered features and trained a robust model under strict time constraints without using guided templates.

• Achieved >88% overall accuracy on the hidden test set, clearly outperforming the baseline.
• Techniques: Ensemble, TTA, image preprocessing, data augmentation
• Secured bonus placement based on model performance metrics.

Re-derived worst-case regret bounds for the Upper Confidence Bound and Linear Upper Confidence Bound algorithms.

• Regret proof for the (Linear) Upper Confidence Bound algorithm, showing sub-linear regret.
• Empirical evaluation on the MovieLens 1M dataset using rejection sampling.
• Graded 1.0.

Trained, fine-tuned, and deployed Tiny YOLOv2 for offline person detection on a Jetson Nano.

• Pre-trained on ImageNet.
• Fine-tuned using an ablation method with progressive unfreezing of layers to preserve low-level features.
• Optimization using int8 and float16 wuantization with per-tensor quantization.
• Graded 1.0.

Extract tables from images and PDFs using PaddleOCR and Mistral OCR.

• Complete user and account management.
• A local extraction pipeline that can run on server CPU.
• Mistral OCR for more complex tasks.
• https://tablexai.de

Master's degree in Computer Science with a focus on ML, data science, and statistics.

Developed novel contextual multi-armed bandit algorithms under fixed budget constraints.

• Integrated neural networks, Gaussian processes, and ridge regression into the decision-making loop.
• Graded 1.0.

Data-driven analysis with a focus on the automotive sector.

• Developed automated web scraping pipelines.
• Built interactive dashboards for sentiment analysis.
• Active market analysis and research.
• First 7 month internship followed by 5 months as a working student.

Review paper about impacts of software on the environment focusing on making software development more environmentally friendly.

• Overview of negative impacts of software on the environment.
• Summary of frameworks for efficient software development, efficient cloud design, cyber-foraging, sustainable reference architectures and Green IT.
• Graded 1.7.

Co-founded and developed Fluxio, a web-based statechart editor designed to simplify robot programming for non-experts.

Bachelor's degree in Computer Science with a focus on theory, ML, and math.