Dimitrios Sarris

o Successfully deployed a XGBoost predictive model optimising discounts for a retailer. Achieved a forecasting error of 4% during Black Friday and Cyber Monday promotions o Developed and productionised a Next Best Message model for a major UK fashion retailer. Generated 9% uplift in accepted demand during A/B tests o Served as a platform engineer/architect for Onsite AI, a Deloitte solution for insurance underwriters powered by Google Cloud

Worked in two research programs under the supervision of Prof. Assimakopoulos o Improved model selection methods for complex time-series forecasting of Federal Reserve Economic Data through bootstrapping techniques by reducing SMAPE error metric by 0.2 to 0.8 %, depending on the selection method (considered an outstanding result for unpredictable discrete-time data). Published proposed methodology as the main author at Operation Research - An International Journal (ORIJ). o Implemented out-of-sample tests for the selection of neural networks for forthcoming M4 forecasting competition based on a benchmark of extrapolative models for analysis of micro and macro financial data among others, resulting in average error reduction per model of 0.5 to 1%