At Urbantz, I built a three-phase XGBoost framework to predict driver route deviations and optimise last-mile delivery sequences. I engineered features from GPS traces, timestamps and operational data, then validated the models with time-based splits and calibration analysis.
I also implemented a greedy local-search algorithm informed by learned driver preferences, cutting predicted deviation risk on 96.5% of candidate routes. At the Port of Piraeus, I analysed customs and trade data and produced compliance reports and trend summaries.
My work spans applied machine learning and business analytics, from building an options-hedging feature at OpenBB to co-building an LLM pipeline for The Circle, an urban loneliness project. I use Python and SQL to turn analysis into recommendations business teams can act on.
