Bridging the gap between cutting-edge research and real-world impact
For my full CV, see my portfolio page or my linktree for a summary of links to my other online profiles.
I'm a multidisciplinary software engineer and computational scientist at the intersection of Computer Vision, Machine Learning, and Bioscience Engineering, with 10 years of combined experience across industry, R&D, and academic environments.
11 years of experience across 5 industry and research environments. Full details under Background below.
| Years | Area | Role & highlights |
|---|---|---|
| 2015–2017 | 🧠 AI in Neurophysiology Lab research |
Deep Learning Researcher (PhD), KU Leuven · Modelling of neurophysiological data using deep convnets |
| 2017–2018 | 🚀 Industry AI Applied solutions |
Data Scientist, Faktion · Computer vision & signal processing; won hackathon (activity recognition in camera feeds) |
| 2018–2024 | 🐞 Agricultural AI Field applications |
PhD + Postdoc, KU Leuven · Built a production-grade, AWS-hosted app for insect identification (annotation, classification & model serving), using computer vision & optical signal processing on edge |
| 2024–2026 | 🛰️ AI for Earth Observation Global monitoring |
R&D Engineer, VITO · Deep learning models for global EO tasks (cloud segmentation & land-cover classification) |
| 2026– | 💽 AI for EO data compression Edge solutions |
R&D Engineer, VITO · Optimization of deep architectures (VQVAE2) for image compression and embedded solutions |
🛰️ R&D Engineer at VITO Remote Sensing
Currently, I'm optimizing VITO's CORSA model (lightweight satellite image compression on edge devices) in the AGES (Applied GeoAI and Edge Solutions) team of VITO's Remote Sensing unit. In parallel, I'm designing custom MLOps tools for internal use, supporting experiment tracking and GPU cluster (on-prem) job management.
📡 Satellite Images → 🔄 CORSA Compression (Edge) → 🖥️ On-Prem GPU Cluster
↳ 🔧 MLOps Tools: Experiment Tracking & Job Management
Before joining AGES, I was part of the RSA team at VITO, developing AI models for the EU's Copernicus Land Cover and Forest Monitoring (LCFM) project (€11M budget). Our team had previously delivered the ESA WorldCover products. I built multi-stage AI pipelines to process satellite data, producing global land cover maps at 10m resolution—ten times sharper than before. My focus was on cloud segmentation models for quality signal identification, maintaining code repositories (soon to be open-sourced), and deploying classification models that generate annual land cover maps, directly supporting EU environmental policy and climate monitoring.
🛰️ Satellite Data → ☁️ AI Cloud Detection → 📊 Quality Composites → 🗺️ Global Land Cover Maps
10m Resolution • Annual Updates • Planetary Scale
In the summer of 2026 I took part in ECMWF Code for Earth, an innovation programme in which developer teams work with ECMWF mentors on open-source Earth-science software. Together with Stylianos Lagaras, I worked on a scalable STAC/Zarr pipeline for ML-ready, multi-source flood inundation observations. The coding phase ran from May to August 2026.
🌊 Project details & mentors
ML-ready archive of satellite-derived flood inundation observations
- Harmonises flood observations from VIIRS (optical), MODIS (optical) and GFM (Sentinel-1 SAR) onto a common 1-arcmin grid.
- Offers a CLI, a Python API and a consolidated Zarr/STAC archive for ML workflows, with a worked example on the 2024 Valencia floods.
🙏 Mentors, with thanks for their guidance throughout:
- Gianpaolo Balsamo (ECMWF, ERA5 and land-surface modelling)
- Calum Baugh (ECMWF, hydrology)
- Kenza Tazi (ECMWF, ML)
- Andreas Grafberger (ECMWF, DevOps)
🛰️ VIIRS + MODIS + GFM → 🧮 Harmonised 1-arcmin grid → 🗄️ Zarr/STAC archive → 🤖 ML-ready flood data
Keywords: Web Development, CLI Tools, DevOps, Computer Vision, Image Processing, IoT, Python, Flask, NiceGUI, Solara, Streamlit
desto - Web dashboard and CLI for managing scripts in tmux sessions
- Full-stack web application with real-time system monitoring, live log viewing, script scheduling, and both web interface and command-line functionality.
vresto - Python interface for Copernicus Sentinel data discovery
- Elegant interface featuring an interactive map for visual searching and high-resolution band visualization.
- Supports dual backends (OData/STAC), granular download management, and a professional CLI/API for programmatic satellite data retrieval.
filoma - File organization & management automation
- Intelligent system designed to categorize and clean up large-scale research datasets based on metadata and content.
- Optimizes pre-processing workflows for ML environments by streamlining directory structures.
plakakia - Python image tiling library for computer vision tasks
- High-performance image tiling tool for object detection and segmentation, utilizing multiprocessing and numpy for efficient processing.
Home_Surveillance_with_Python - Motion detection surveillance system
- End-to-end Edge AI product featuring real-time motion analysis on Raspberry Pi with cloud-notified alerts.
AI/ML: Computer Vision • Deep Learning • CNNs • YOLO • Time-Series Analysis
Cloud: AWS • Docker • FastAPI • Web GUIs (Streamlit, Solara, NiceGUI...)
Data: Hyperspectral Imaging • Satellite Data • IoT Sensors • Big Data Processing
🛰️ Remote Sensing & AI at VITO (Current; 2024-)
- AGES team (2026-): Optimize CORSA, VITO's lightweight satellite image compression model for edge devices, and design custom MLOps tools for experiment tracking and on-prem GPU cluster job management
- RSA team (2024-2026): Developed reliable earth land cover classification systems through LCFM project
- Apply hyperspectral satellite data analysis and ML models for environmental datasets
- Work with cloud services, Hadoop, Spark, and AWS for large-scale processing
- Contribute to EU Commission projects for sustainable development and climate change mitigation
🐞 Bioscience Engineering at KU Leuven (PhD & Postdoc; 4 + 2 years; 2018-2022 + 2022-2024)
PhD Focus: Optical Insect Identification using AI
Postdoc: Productized research by developing a production-ready API and web dashboard, bridging the gap between raw AI models and end-user accessibility. Led AI projects, mentored PhD researchers, and specialized in hyperspectral imaging.
- Deployed production API server on AWS serving AI models and tools to external companies via a web interface (demo-setup, demo-interface, demo-label-tool)
- Built comprehensive AI systems for insect recognition using computer vision and audio analysis (see insect-trap, wbai, stickybugs-ai)
- Developed software tools for IoT devices, desktop systems, and cloud infrastructure (see photobox, example publication)
Created the plakakia library | 5 publications in high-impact journals
🚀 Data Scientist at Faktion (1 year; 2017-2018)
Applied AI solutions for industry clients including predictive maintenance (Bridgestone), sales analytics (Aliaxis), and computer vision POCs.
Achievement: 🏆 Won hackathon on Activity Recognition (Vinci Energies)
🧠 Deep Learning Research at KU Leuven (2 years; 2015-2017)
Studied deep CNNs and their resemblance to biological visual systems. Developed models to predict neuronal activity from artificial neuron activations.
Specialized in the deep learning components of all published research | 4 publications in top neuroscience journals | Presented my work at VSS conference (Florida, USA). List of papers below:
-
The ventral visual pathway represents animal appearance over animacy, unlike human behavior and deep neural networks. Bracci, S., Ritchie, J. B., Kalfas, I., & Op de Beeck, H. P. (2019). The Journal of Neuroscience, 39(33), 6513-6525. DOI
-
Representations of regular and irregular shapes by deep Convolutional Neural Networks, monkey inferotemporal neurons and human judgments. Kalfas, I., Vinken, K., & Vogels, R. (2018). PLOS Computational Biology, 14(10), e1006557. DOI
-
Representation of semantic similarity in the left intraparietal sulcus: functional magnetic resonance imaging evidence. Neyens, V., Bruffaerts, R., Liuzzi, A. G., Kalfas, I., Peeters, R., Keuleers, E., Vogels, R., De Deyne, S., Storms, G., Dupont, P., & Vandenberghe, R. (2017). Frontiers in Human Neuroscience, 11, 402. DOI
-
Shape selectivity of middle superior temporal sulcus body patch neurons. Kalfas, I., Kumar, S., & Vogels, R. (2017). eNeuro, 4(3). DOI
🐞 PhD in Bioscience Engineering (KU Leuven, Belgium 🇧🇪; 4 years; 2018-2022)
Successfully completed doctoral research with a focus on optical insect identification using signal processing and computer vision with AI/ML/DL
🧠 PhD Research* - Neurophysiology Lab (KU Leuven, Belgium 🇧🇪; 2 years; 2015-2017)
*Fulfilled requirements early, but exited programme
Explored computational neuroscience applications and deep learning models for biological neurons
🎓 MSc. Machine Learning (KTH Royal Institute of Technology, Stockholm, Sweden 🇸🇪; 2015)
Specialized in Computational Neuroscience and Spiking Neural Networks
Thesis research simulating neocortical structures using NEST simulator in Python
🎓 BSc. Computer Science (Aristotle University of Thessaloniki, Greece 🇬🇷; 2013)
Built a solid foundation in computing theory and educational information systems
🌱 I'm always interested to learn about how Artificial Intelligence can improve our lives.
💬 To reach out, send an email at kalfasyan[at]gmail[dot]com
🔗 Check my linktr.ee
📚 Researcher profiles:
🧬 orcid
🔬 scholar
📖 researchgate
🌐 Stay connected through the following social media channels: bluesky, linkedin, github