Recent Projects
For an overview of my earlier and non-technical projects, please visit my LinkedIn profile.
Academic
Modelling Week-To-Week Voting Intention Change (2024 – 2025)
Master’s thesis project, M.Sc. Data Science for Public PolicyDetails
Context Inter-election vote switching is typically based on recall items and/or election results. The thesis aimed to develop a method for estimating transition matrices of short-term party preference change in a higher temporal resolution using high-frequency panel data.
Data SOSEC panel (n=3500, bi-weekly, USA/Germany, 2023-2025), FZI
Methods Hierarchical Bayesian imputation model in the latent k-dimensional simplex space
Results Bi-weekly voting intention transition matrices, run-up to the 2025 German federal election
Building a Shiny Web App for Ternary Plots of Generalised Dirichlet Distributions (2025)
Side project to the Master’s thesis, M.Sc. Data Science for Public PolicyDetails
Context There is a lack of interactive modelling helper tools when working with (generalised, that is, mixtures of) Dirichlet distributions. The web app aims to fill this gap.
Methods Simulation of random draws, visualisations in a R/Shiny app hosted on Posit Connect
Building an LDA Classifier Model for Full Bayesian Text Classification with STAN (2024)
Seminar project, M.Sc. Data Science for Public PolicyDetails
Context Many NLP classifiers are implemented in a frequentist/ML fashion. The project aimed to implement a full Bayesian classifier model using Latent Dirichlet Allocation (LDA).
Data Synthetic text data based on Telegram messages from a Berlin public transport group
Methods Inference of LDA parameters with STAN, posterior prediction, model evaluation
Building a Joint Latent Model of Media Coverage Indicators and Issue Salience (2024)
Seminar project, M.Sc. Data Science for Public PolicyDetails
Context As the climate movement became popular in 2019, the German’s perception of most important problem (MIP) shifted towards climate change. It was the aim of this project to combine different data sources on media coverage and issue salience polling data in a joint model.
Data News coverage data, SPIEGEL & GDELT; MIP item, Forschungsgruppe Wahlen (2019)
Methods Full Bayesian joint latent regression model with STAN
Results Quantification of a news coverage effect of Fridays For Future protests on climate change issue salience
Analysing the Sudden Surge in Climate Awareness in Germany 2019 (2023)
Master’s thesis project, M.P.P. Master of Public PolicyDetails
Context As the climate movement became popular in 2019, the German’s perception of most important problem (MIP) shifted towards climate change and the Green party’s popularity surged. The thesis aimed to analyse the multi-faceted effects Fridays For Future’s protests had.
Data MIP and voting intention items, Forsa & Forschungsgruppe Wahlen (2019); protest records, Fridays For Future Germany
Methods Theory development, regression analysis
Results Quantification of a news coverage effect of Fridays For Future protests on climate change issue salience
Assessing the PRC’s Climate Responsibility Using GHG Emissions Data (2022)
Seminar group project, M.P.P. Master of Public PolicyDetails
Context It is a matter of perspective (historic vs. current emissions) what the PRC’s fair share of global climate mitigation efforts should be. The project aimed to shed light on the reduction pathways the PRC could take to reach climate neutrality.
Data GHG emissions data (1920-2020, EDGAR); population and economic data (1990-2020, World Bank)
Methods Descriptive analysis, visualisations, transition path forecasting models
Results Quantification of the PRC’s historical, current, and future emissions contributions
Forecasting Germany’s Demand for Natural Gas in the Energy Crisis (2022)
Seminar group project, M.P.P. Master of Public PolicyDetails
Context In the wake of the energy crisis following Russia’s invasion of Ukraine, Germany’s natural gas prices surged. The project aimed to develop forecast models for real-time demand prediction.
Data Various data (2017-2022): historic gas prices and demand (THE), weather data (Meteostat), CO2 prices (investing.com), electricity prices (investing.com), DAX stock market index (Yahoo finance), German GDP quarterly (World Bank)
Methods Machine learning time series models
Results Real-time week-ahead forecast of national natural gas demand
Professional
Developing a Post-Stratification and Imputation Framework for the SOSEC Panel (2025)
Methodological project for FZI Forschungszentrum Informatik, BerlinDetails
Context With high panel mortality, unbalanceness and inconsistencies in the sampling process, the SOSEC panel requires robust post-stratification and imputation methods to reduce bias and increase representativity.
Data Census data (Germany: DeStatis, USA: US Census Bureau); SOSEC panel (n=3500, bi-weekly, USA/Germany, 2023-2025), FZI
Methods Post-stratification weights (joint distributions of demographic variables); multiple imputation
Analysing the Spatial Distribution of Balcony Solar Panels in Berlin (2024)
Data analysis project for PLAN B 2030 e.V., BerlinDetails
Context State-level funding programmes have boosted the installation of small-scale solar panels in Berlin. The project aimed to analyse the spatial distribution of these panels and to draw conclusions on the success of the programme.
Data Marktstammdatenregister (2021-2023, BNetzA)
Methods Geospatial data analysis, visualisations
Results There is a surprising concentration of installations in wealthier single family house areas, with indications for the funding programme.
Building a Web App for a Unified Postal Vote Registration for German Elections (2023)
Professional Year project for Mehr Demokratie e.V.Details
Context Online postal vote registration, as many other citizen services in Germany, is in the responsibility of the municipalities. The project aimed to build a unified web app to simplify the process.
Data Federal and state-level statistical offices, ZIP code data (Deutsche Post)
Methods Matching of addresses, zip codes, administrative boundaries; web development: landing page, template-based application generator, email reminder service, information portal
Personal
Building a Risk Model of Berlin’s Public Transport Ticket Inspections (2024)
Not-for-profit app development for Freifahren BerlinDetails
Context A telegram group with more than 25k members reports on sightings of ticket inspection staff in Berlin. The project aimed to develop a risk model of getting checked in Berlin’s public transport.
Data Telegram group data, Overpass Turbo (OSM)
Methods Geospatial data analysis, risk modelling
Thanks for your interest!
Did you know I am currently looking for a PhD position? Also see my Academic CV.
Please feel free to reach out to me or connect through LinkedIn.