Project Title: Nationwide Automated Land Cover Mapping System

Responsibilities:

• Vision-Centric MLOps (Human-in-the-Loop): Designed and managed an end-to-end active learning pipeline; transformed model inferences into vector geometries for expert review in a database, subsequently converting corrected vectors back to raster masks for iterative model retraining.
• State-of-the-Art (SOTA) Implementation: Developed a large-scale land cover mapping system for Greece using VHR imagery; integrated DINO-based feature extraction and SAM (Segment Anything Model) for precise boundary refinement and semantic segmentation.
• Advanced Remote Sensing Preprocessing: Engineered high-performance workflows including MAJA atmospheric correction, Pansharpening, and Super-resolution; implemented artifact-free stitching and tiling to ensure seamless nationwide data consistency.
• Bio-Physical Modeling: Integrated Canopy Height Estimation models utilizing spectral and geometric features to deliver multi-layered environmental insights beyond standard classification.

Strategic Partners: Ministry of Digital Governance (Greece)

Project Title: Novel Edge-AI system for accurate and near real-time plastic detection and monitoring in marine environment

Responsibilities:

• Space-Edge Architecture Design: Engineered an ultralight U-Net architecture for on-board marine plastic detection; optimized the model to fit within the extreme power and memory constraints of orbital hardware.
• Model Compression & Optimization: Developed pruning, transfer learning, and quantization workflows to enable real-time inference on low-power chips; maintained high radiometric accuracy for multispectral data while meeting strict computational latency targets.

Strategic Partners: Edge SpAIce consortium; ESA and Horizon Europe project consortia; NCSR "Demokritos"

Project Title: An Intelligent Early-Warning Oil Spill Detection and Prediction System for the Arabian Gulf and the Red Sea

Responsibilities:

• Autonomous End-to-End Pipelines: Architected, developed, and deployed automated, Near-Real-Time (NRT) MLOps pipelines (preprocessing, inference, final analytics, and insights) for oil spill detection, integrating multi-sensor data (SAR and Optical) into containerized (Docker) production environments.
• Satellite Data Preprocessing: Engineered automated Python workflows for multispectral and SAR data; implemented Sen2Cor and Acolite corrections alongside custom SNAP/Snappy wrappers for SAR GRD-to-Sigma0 processing to ensure highfidelity Analysis-Ready Data (ARD).
• Custom Dataset & Architecture Design: Created multi-sensor large-scale geospatial datasets; designed and trained lightweight semantic segmentation architectures using hardware-specific optimizations to maximize inference speed for NRT results.
• Strategic Technical Leadership: Acted as primary technical lead, translating complex requirements into well-scoped experiments, CI/CD workflows, and automated QA tools; enforced high standards for clean, typed Python code and rigorous peer code reviews.
• Advanced Analytics & Visualization: Developed custom internal tooling for spectral analysis and spatial visualization, transforming raw model outputs into actionable georeferenced insights for environmental decision-making.

Strategic Partners: King Abdullah University of Science and Technology (KAUST); Saudi Aramco-KAUST Marine Environmental Observations Center (SAKMEO)