I build analysis that goes from raw data to a decision. My projects run end to end: ingestion, modelling, validation, and an interactive front end to explore the results. Most of them work on EU regional and macro data, using causal and panel methods.
|
Panel econometrics of R&D and productivity across 29 European countries, 1998–2024. Covers GMM/IV, spatial spillovers, local projections and causal forests. Extends my bachelor's thesis.
|
Causal analysis of Poland's 1999 reform that demoted 31 cities from province capitals. Six estimators agree on direction: synthetic DiD puts the loss at ~4.5% of population.
|
|
Graph neural network autoencoder that scores economic and political anomalies across EU NUTS2 regions. Unsupervised, and it recovers the 2020 COVID shock without being told the date.
|
Interactive four-layer map dashboard of EU NUTS2 regions: reform impact, innovation archetypes, investment suitability and GDP per capita.
|
|
Desktop app that reproduces my bachelor's thesis: augmented Solow model, K-Means typology, FE/RE with Hausman, unit-root tests and causal designs. Ships with CI and a Windows build.
|
KMeans, GMM and hierarchical clustering compared on silhouette, ARI and run-to-run stability. Finding: initialization mattered more than algorithm choice.
|
More: ro-administrative-reform · eu-innovation-panel · eu-megacampus-siting · ro-voting-prediction · dsk805-companion
| Econometrics and causal inference | Panel FE/RE, GMM/IV, difference-in-differences, synthetic control, local projections, panel VAR |
| Machine learning | scikit-learn, PyTorch, graph neural networks, clustering, causal forests |
| Data and visualization | pandas, DuckDB, matplotlib, Streamlit, Next.js, React |
| Engineering | Git, GitHub Actions, reproducible pipelines |
