At Wikimedia Foundation, I extended Python tools that create and populate Wikidata entries for Wikipedia biographies. I expanded detection from two languages to five and added Spanish and Italian date extraction.
At Instituto de Ciencias Físicas, I built reproducible Python pipelines for Nikkei 225 market data. Time-series analysis, hypothesis tests, and rolling-correlation plots helped identify structural changes in how financial assets moved together.
In my Crowd Forecast Evaluation project, I tested whether nine pooling rules transferred between crowds. Pushing forecasts away from 50% improved Brier scores on later questions but worsened them for ForecastBench's public crowd.
My other projects examine fraud policy, inventory replenishment, and retail operations. I built a buy-now-pay-later fraud simulator, evaluated replenishment policies under a storage limit, and created a Power BI retail sales dashboard.
