Resume
Profile
I’m Fabian Rosenthal, a Data Scientist and developer. With my engineering background, I bring a cross-disciplinary perspective to data science. I am specialized in performance estimation and uncertainty quantification. In my work and university projects, I have proven the ability to identify the right tool for a problem, communicate it to the team, master it, and ship it quickly.
Languages
- English (C1)
- German (native)
Soft skills
- Meticulous
- Versatile
- Empathetic
- Creative
Technical Skills
Programming
- Python, R, SQL
- LaTeX, Matlab
- Rust (learning)
Frameworks
- PyTorch, scikit-learn
- Pandas, Polars, Arrow
- mlr3, data.table
- tidyverse, tidymodels
- Bokeh, Shap, ggplot2
Tools
- Quarto, Tableau
- Git, Docker, Linux CLI, Parquet
Experience
Student Research Assistant
Fraunhofer MEVIS (Bremen) | 02/2024 – 08/2024
- Developed statistical R-packages with uncertainty quantification techniques, giving clinicians a more robust understanding of ML models.
- Designed unified APIs with consistent code styling and good unit tests.
Student Research Assistant
University of Applied Sciences Düsseldorf | 09/2020 – 10/2023
- Co-published a data set with soundscapes from 100+ participants.
- Performed data and feature engineering on 6000+ audio recordings.
- Saved days of compute by employing a faster hyperparameter tuning.
- Introduced a solution to keep track of 1000+ experiments.
- Keynotes and posters on the use of stats & ML in acoustic research.
Video Recording Producer
WDR (Köln) | 09/2020 – 02/2022
- Called cameras for live-streaming concerts in a musically sensible way.
- Led technical teams and communicated cinematographic ideas efficiently.
Bike Messenger
Self-employed (Düsseldorf) | 11/2016 – 09/2020
Completed time-critical and valuable deliveries in all weather conditions while optimizing complex urban routes in real-time.
Projects
flexcv
University of Applied Sciences Düsseldorf & Private Project
- A Python package simplifying nested cross-validation on grouped data.
- Enables reproducible multimodel workflows with few lines of code.
- Released and maintained as OSS using GitHub Actions (CI/CD).
- Project page
Time Series Forecasting
University of Applied Sciences Cologne
- Prediction of open bicycle counter data for Cologne and benchmarking of ML-algorithms with time series cross-validation.
- Fine-tuned XGBoost with external weather data and seasonal trends beats the best linear models.
GridNet White Balance
University of Applied Sciences Cologne
- Achieved improvement in color consistency metrics with GridNet.
- Deployed the model within a custom camera pipeline.
- Strong communication and co-operation in a 5-person team.
Soundscape Viz App
University of Applied Sciences Cologne - Interactive app visualizing survey data on individual levels. - Enables intuitive exploration of the diversity of participants. - Deployed here on the Posit Connect Cloud.
Education
Master of Science: Media Technology
University of Applied Sciences Cologne | 2022 – 2024
- Thesis: Comparison of uncertainty quantification techniques for empirical prediction performance based on cross-validation. Optimized distributed ML pipeline to train 4.3M+ models on the Fraunhofer Edge Cloud. A writeup is available here.
- Courses: Machine learning; Deep learning and object recognition; Data visualization; Multivariate statistics.
- Final grade: 1.2
Bachelor of Engineering: Sound and Video
RSH & University of Applied Sciences Düsseldorf | 2010 – 2022
- Thesis: Audio feature extraction for predicting indoor soundscapes.
- Final grade: 1.5
Other Accomplishments
On the podium: Played solo horn in Bruckner’s 8th symphony with the Düsseldorf University Orchestra.
On two wheels: Cycled 300 km self-supported through Hessen.