Vasileios is currently unavailable
Vasileios Athanasiou
Verified Expert in Engineering
Artificial Intelligence Engineer and Developer
Chalcis, Euboea, Greece
Toptal member since January 30, 2026
Bio
Vasileios is an AI and ML engineer with a PhD and a track record of translating complex R&D into production-grade systems. He is an expert in developing algorithms for both vision and sensor data, bridging the gap between state-of-the-art academic research and industrial deployment. Vasileios has also developed scalable AI and ML solutions and is dedicated to leading technical initiatives and translating complex theories into production-grade applications.
Portfolio
Kela
C++, Computer Vision, Python, CUDA Kernel, OpenCV, DeepStream SDK...
inos Automationssoftware GmbH
C++, Python, ONNX Runtime, Open Neural Network Exchange (ONNX), Point Clouds...
DeepPath
Computer Vision, Deep Learning, PyTorch, Decision Trees...
Experience
- Python - 8 years
- Deep Learning - 7 years
- PyTorch - 6 years
- Computer Vision - 6 years
- OpenCV - 5 years
- Object Detection - 4 years
- C++ - 3 years
- Vision Transformer (ViT) - 2 years
Preferred Environment
PyTorch, C++, OpenCV, Computer Vision, Linux, Windows, Visual Studio Code (VS Code), Docker, Artificial Intelligence (AI), Python
The most amazing...
...work I've done is architect and deploy an efficient real-time computer vision pipeline at the edge in C++, specializing in efficient CUDA optimization.
Work Experience
Senior Computer Vision Engineer
2025 - 2025
Kela
- Engineered an end-to-end, low-latency video stabilization pipeline in C++ for processing real-time streaming protocols integrated within the NVIDIA DeepStream framework.
- Implemented the core motion estimation using a custom optical flow algorithm, directly accelerating it with the OpenCV CUDA library and CUDA Streams. Conducted rigorous performance profiling with NVIDIA Nsight to achieve maximum GPU throughput.
- Engineered core multi-view geometry algorithms, including robust pipelines for camera intrinsic and extrinsic calibration and 3D triangulation to solve complex spatial mapping challenges.
- Leveraged Kalman filters for robust state estimation, creating a temporal smoothing layer that filtered noise and produced stable, jitter-free estimations of system state over time.
- Containerized the entire perception stack using Docker, ensuring robust, reproducible deployment into the company’s operational pipelines.
Technologies: C++, Computer Vision, Python, CUDA Kernel, OpenCV, DeepStream SDK, NVIDIA Nsight Systems, Linux, CMake, GStreamer, Embedded C++, NVIDIA CUDA, GitHub, Engineering Software, NumPy, SciPy
Senior Computer Vision Engineer
2023 - 2025
inos Automationssoftware GmbH
- Engineered and deployed real-time computer vision models as hybrids: computer vision algorithms and deep learning models. Applications include 2D/3D defect detection and part-traceability via optical character recognition (OCR).
- Transitioned these models from Python research prototypes into low-latency C++ inference pipelines, optimized for the production environment using tools like ONNX.
- Developed a proprietary 2D/3D AI infrastructure toolset in C++, including custom annotation tools. Clients can instantly provide feedback and generate new datasets for model training.
Technologies: C++, Python, ONNX Runtime, Open Neural Network Exchange (ONNX), Point Clouds, Computer Vision, OpenCV, Deep Learning, Visual Studio, PyTorch, Convolutional Neural Networks (CNNs), SQL, Object Detection, Semantic Segmentation, Embedded C++, GitHub, Engineering Software, NumPy, Pandas, SciPy
Senior AI Scientist
2022 - 2023
DeepPath
- Conducted advanced industrial R&D in medical imaging.
- Developed novel deep learning architectures for high-precision diagnosis in large histopathological images. Designed feature-embedding post-processing networks that improved diagnosis.
- Developed deep learning segmentation pipelines for accurately segmenting biological objects in 2D.
- Implemented object tracking in 2D with deep learning for efficiently estimating their trajectories despite not being detected.
Technologies: Computer Vision, Deep Learning, PyTorch, Decision Trees, Vision Transformer (ViT), Docker, Bash, Linux, Python, Tracking, Embedding Models, Object Detection, Semantic Segmentation, Convolutional Neural Networks (CNNs), XGBoost, Detectron, NVIDIA CUDA, GitHub, NumPy, Pandas, SciPy
Postdoctoral Researcher, Max Planck Institute for Dynamics of Complex Technical Systems
2021 - 2022
Max-Planck-Gesellschaft
- Pioneered the use of physics-informed machine learning (PIML) to model dynamical systems.
- Engineered novel deep learning architectures that integrated physical laws into the training process.
- Applied advanced feature engineering to extract state variables from high-dimensional data with a physically meaningful representation of the system.
Technologies: PyTorch, Python, Jupyter Notebook, PyCharm, GitHub, NumPy, Simulations
PhD Researcher
2016 - 2020
Chalmers University of Technology
- Developed novel machine learning algorithms based on reservoir computing for the classification of complex temporal data sequences.
- Simulated memory-resistor-based neural networks, contributing to the field of neuromorphic computing or hardware-aware AI.
- Built extensive numerical simulation frameworks using Python and Java, adhering to strict object-oriented programming principles to ensure code scalability.
- Collaborated with experimental hardware groups, validating theoretical models against real-world physical measurements.
- Authored multiple scientific publications in high-impact journals, presenting novel theories on nonlinear dynamical systems.
Technologies: Python, Java, Regression, Numerical Simulations, NumPy, Simulations
Experience
Video Stabilization Pipeline with NVIDIA DeepStream
An end-to-end, low-latency video stabilization pipeline in C++ for processing real-time streaming protocols, integrated within the NVIDIA DeepStream framework.
• Implemented the core motion estimation using a custom optical flow algorithm, directly accelerating it with the OpenCV CUDA library and CUDA Streams.
• Conducted rigorous performance profiling with NVIDIA Nsight to achieve maximum GPU throughput.
• Leveraged Kalman filters for robust state estimation, creating a temporal smoothing layer that filtered noise and produced stable, jitter-free estimations of system state over time.
• Containerized the entire perception stack using Docker, ensuring robust, reproducible deployment into the company’s operational pipelines.
• Implemented the core motion estimation using a custom optical flow algorithm, directly accelerating it with the OpenCV CUDA library and CUDA Streams.
• Conducted rigorous performance profiling with NVIDIA Nsight to achieve maximum GPU throughput.
• Leveraged Kalman filters for robust state estimation, creating a temporal smoothing layer that filtered noise and produced stable, jitter-free estimations of system state over time.
• Containerized the entire perception stack using Docker, ensuring robust, reproducible deployment into the company’s operational pipelines.
OCR for Part-traceability
An OCR computer vision pipeline combining classical computer vision algorithms with deep learning.
• Implemented and optimized classical computer vision algorithms with OpenCV/C++ for identifying the text lines, rotating the text lines to become axis-aligned, and segmenting the characters of the text lines.
• Trained with PyTorch/Python convolutional neural networks (CNNs) for classifying the segmented characters.
• Deployed the trained CNN with ONNX runtime into the C++ environment.
• Implemented and optimized classical computer vision algorithms with OpenCV/C++ for identifying the text lines, rotating the text lines to become axis-aligned, and segmenting the characters of the text lines.
• Trained with PyTorch/Python convolutional neural networks (CNNs) for classifying the segmented characters.
• Deployed the trained CNN with ONNX runtime into the C++ environment.
3D Point Cloud Classification
Training and deploying a 3D point cloud classifier by combining deep learning features with hand-crafted features.
• Utilized a pre-trained 3D point cloud neural network for deep learning feature extraction.
• Trained XGBoost models for feature selection.
• Trained XGBoost models as classifiers that take as input the selected deep learning and hand-crafted features.
• Deployed the whole network in ONNX runtime/C++. The whole network consists of a point cloud feature extractor followed by XGBoost models.
• Utilized a pre-trained 3D point cloud neural network for deep learning feature extraction.
• Trained XGBoost models for feature selection.
• Trained XGBoost models as classifiers that take as input the selected deep learning and hand-crafted features.
• Deployed the whole network in ONNX runtime/C++. The whole network consists of a point cloud feature extractor followed by XGBoost models.
Accurate Border Segmentation of Biological Objects
Accurate segmentation of biological objects and their borders from microbiological images.
• Implemented classical computer vision algorithms on ground truth masks to extract border masks.
• Trained multi-class semantic segmentation neural networks, such as UNET and DeepLabV3, on the task background vs. internal vs. border.
• Designed post-processing computer vision algorithms to treat the semantic segmentation results and provide predicted 2D objects.
• Provided 2D object detection evaluation metrics.
• Tracked all the experiments with MLflow.
• Communicated with experts for improving data annotations.
• Sped up the data annotation through AI. Used existing models for providing initial data annotations.
• Implemented classical computer vision algorithms on ground truth masks to extract border masks.
• Trained multi-class semantic segmentation neural networks, such as UNET and DeepLabV3, on the task background vs. internal vs. border.
• Designed post-processing computer vision algorithms to treat the semantic segmentation results and provide predicted 2D objects.
• Provided 2D object detection evaluation metrics.
• Tracked all the experiments with MLflow.
• Communicated with experts for improving data annotations.
• Sped up the data annotation through AI. Used existing models for providing initial data annotations.
Education
2016 - 2020
PhD in Dynamical Systems, Numerical Simulations, Machine Learning, and Reservoir Computing
Chalmers University of Technology - Gothenburg, Sweden
2012 - 2014
Master's Degree in Biomedical Engineering
Linkoping University - Linkoping, Sweden
2006 - 2011
Engineer's Degree in Computer Engineering
National Technical University of Athens - Athens, Greece
Skills
Libraries/APIs
PyTorch, OpenCV, Scikit-learn, TensorFlow, NumPy, Pandas, SciPy, XGBoost
Tools
PyCharm, Open Neural Network Exchange (ONNX), Visual Studio, GitHub, Mathematica, MATLAB, NVIDIA Nsight Systems, CMake
Languages
C++, Python 3, Python, Embedded C++, Java, Bash, SQL
Platforms
Linux, Windows, Visual Studio Code (VS Code), Jupyter Notebook, NVIDIA CUDA, Docker
Frameworks
GStreamer
Other
Computer Vision, Deep Learning, Artificial Intelligence (AI), Digital Signal Processing, Medical Imaging, Mathematical Modeling, Numerical Simulations, Dynamic Systems, Decision Trees, Vision Transformer (ViT), Tracking, Embedding Models, Regression, Point Clouds, Object Detection, Semantic Segmentation, Simulations, Electrical Circuits, ONNX Runtime, CUDA Kernel, DeepStream SDK, Engineering Software, Convolutional Neural Networks (CNNs), Detectron, Finite Element Analysis (FEA)
Collaboration That Works
How to Work with Toptal
Toptal matches you directly with global industry experts from our network in hours—not weeks or months.
1
Share your needs
Discuss your requirements and refine your scope in a call with a Toptal domain expert.
2
Choose your talent
Get a short list of expertly matched talent within 24 hours to review, interview, and choose from.
3
Start your risk-free talent trial
Work with your chosen talent on a trial basis for up to two weeks. Pay only if you decide to hire them.
Top talent is in high demand.
Start hiring