Giannis Dravilas

• Forecast Benchmarking (FACT): Designed and developed FACT (Forecast Accuracy Tracking), a system for benchmarking numerical weather prediction models against observations from 8,000+ weather stations, supporting forecast evaluation and model selection.
• Hyperlocal Forecasting (MOSAIC): Developed WeatherXM MOSAIC, a hyperlocal forecasting system that selects weather models based on their recent performance at individual stations. MOSAIC ranked first for one-day-ahead temperature accuracy in FACT's 2025 network-wide evaluation (1.37 °C MAE).
• Data Quality (QoD): Refactored and optimised WeatherXM's Python-based Quality of Data (QoD) mechanisms for meteorological time series, achieving a 50% improvement in processing speed. These mechanisms support weather station data validation.
• Dashboards & Visualisation: Built Grafana dashboards and visualisations to communicate forecast performance, data quality and operational insights to technical and non-technical stakeholders.
• Scientific Software Engineering: Delivered modular, documented and tested Python code, including AWS S3 integration, pytest-based tests and code reviews.
• Collaboration & Mentorship: Collaborated closely with a meteorologist and stakeholders on requirements, evaluation and interpretation, while mentoring a junior data scientist on scientific software engineering practices.

• Python
• SQL
• NetCDF
• GRIB2
• Parquet
• xarray
• Docker
• Git
• S3
• AWS Athena
• Grafana
• Meteorological Data
• Numerical Weather Prediction
• Weather Station Observations
• Timeseries
• Machine Learning

• AI-Ready Data Infrastructure: Worked on AI-ready data infrastructure and machine learning workflows in the context of Destination Earth and DestinE AI-Factories.
• Deep Learning on Satellite Data: Built and trained deep learning models to estimate high-resolution radar reflectivity from Meteosat Third Generation FCI/LI observations, using OPERA radar data for supervision, in collaboration with the Remote Sensing Products Division.
• Large-Scale Data Engineering: Created TB-scale AI-ready datasets in Zarr format and addressed I/O bottlenecks affecting large-scale model training and inference.
• Cloud-Native ML Workflows: Worked with the Destination Earth Data Lake, including S3-based storage and compute infrastructure, to support cloud-native ML experiments using weather and Earth observation data.
• Distributed Training & HPC: Scaled PyTorch Distributed Data Parallel training across 64 A100 GPUs on the EuroHPC MeluXina cluster, achieving near-linear scaling.
• MLOps & Reproducibility: Developed reproducible ML workflows using Docker, Apache Airflow for automated inference, and MLflow for experiment tracking, with GitLab for version control and collaboration.
• Scientific Communication: Contributed to work presented at the EUMETSAT Meteorological Satellite Conference 2026.

• Python
• PyTorch
• Zarr
• HPC
• Distributed Data Parallel
• Docker
• Apache Airflow
• MLflow
• S3
• GitLab
• Deep Learning
• Vision Transformers
• Convolutional Neural Networks
• Remote Sensing
• Satellite Data
• Weather Radar Data

• EuroHyPerCon Project: Contributed as an independent consultant to EuroHyPerCon, a EuroHPC JU-funded initiative focused on connectivity requirements and network design for high-performance computing systems in Europe.
• Unstructured Data Analysis: Analysed unstructured text data to identify trends, requirements and actionable insights related to HPC connectivity and infrastructure planning.
• Visualisation & Reporting: Developed visualisations and summaries to communicate technical findings clearly to project stakeholders. The project's final report was published by the Publications Office of the European Union.

• Climate Risk Assessment: Contributed to early-stage data scoping and methodology design for a Climate Risk Assessment Platform.

• Thesis: SATFRONT: Detecting Atmospheric Fronts from Satellite Data using Deep Learning (arXiv), in collaboration with EUMETSAT.
  • Developed a Swin-UNet to detect cold, warm and occluded fronts from GEO-Ring satellite observations, trained on analyses from five European meteorological services.
  • Built an 18 TB AI-ready GEO-Ring/ERA5 dataset using Zarr and S3 on Destination Earth, plus a web demo producing near-real-time frontal analyses every 30 minutes.

  Supervisors: Miruna Stoicescu, Roope Tervo, Carlos Horn, Oriol Hinojo Comellas, Sina Montazeri, Andrea Meraner (EUMETSAT); Ana Lučić (UvA).

• Research & Publications: An arXiv preprint on atmospheric front detection, two journal submissions on Vision Transformers (1, 2), one Information Retrieval submission (under review), and a Human-Computer Interaction paper presented at IEEE PacificVis 2026.

• Thesis: Machine Learning Snowfall Retrieval Algorithms for Satellite Precipitation Estimates, developed in collaboration with IERSD/NOA.
  • Selected for the department's annual e-book showcasing outstanding theses and resulted in a peer-reviewed publication.

  Supervisors: Manolis Koubarakis (NKUA); Stavros Dafis, Georgios Kyros, Kostas Lagouvardos (NOA).

Founder and owner of Weatherathens and snowly.eu. Nature photography work featured online by research institutions, e.g., National Observatory of Athens.

Translate user interface content and computer science terminology for the Indico project from English to Greek.

Conducted research on climate risk adaptation, communication and decision-making. The work resulted in a publication presented at the EGU General Assembly 2026 (Abstract).

Led a hands-on workshop on data science and ML for weather/climate using Python and key libraries.

Supported first-year students in understanding core concepts and completing lab exercises for the courses "Introduction to Programming" and "Data Structures and Programming Techniques".

Translated user interface content and meteorological terminology from English to Greek.