Panagiotis Zogas

• Participated in the European project DigIt, developing spatiotemporal forecasting models for human trajectories in mines using Kalman filters and probabilistic deep learning models. Conducted trajectory analysis, preprocessing tasks (scaling, augmentation, smoothing), and created simulated data to support an early warning system.
• Performed data interrogation for a travel package client, including hypothesis testing for product comparisons, funnel analysis, saturation point analysis, product list analysis, market basket analysis on sparse data, attribution modeling with rule-based and data driven methods, and A/B testing.
• Designed and implemented stateful Markov chain algorithms for streaming data-driven attribution modeling.
• Developed and implemented Azure ML pipelines for preprocessing, augmentation, and classification of large image datasets using transformer models for image processing at the National Bank of Greece, overseeing model and dataset registration and deployment with MLOps.
• Developed a machine learning framework for time series forecasting within the PLANET4HEALTH European project (OneHealth applications), integrating preprocessing methods, statistical and ML models for both deterministic and probabilistic forecasting, fine-tuning pipelines, creating explainability modules, and applying MLOps practices, all within an automated tool that utilizes simple JSON configurations.
• Developed an ML-powered M&V module within the Prelude European project, using simple regression models combined with bootstrapping to predict electrical energy consumption for the purpose of M&V analysis.
• Supervised and mentored six entry-level data scientists, providing guidance and support throughout their projects and professional development.

During the mandatory military obligations, I was stationed at KEPYES in Athens, where I primarily provided technical support for the information systems at the Operations Headquarters of the Hellenic Army.

• Analyzed product consumption data, focusing on data exploration and time series anomaly detection using statistical methods, clustering methods, deep auto-encoder models, and forecasting models.
• Contributed to a customer segmentation task using RFM analysis, clustering, and random forests to enhance explainability.
• Part of a team that secured 2nd place in the EYDAP Innovation Challenge, designing reproducible pipelines for time series forecasting. Contributions included data exploration, incorporating preprocessing methods (e.g., interpolation, sample alignment), developing statistical and ML models for regular and irregular time series, integrating feature extraction and clustering techniques, and implementing model ensembling.
• Researched and developed an A/B testing module with support for multiple comparisons, incorporating frequentist and Bayesian methods, hypothesis testing with early stopping, and solutions for the multi-armed bandit problem.