In my role as a professional in Medical Imaging Segmentation, I specialize in training advanced neural networks for accurate segmentation of medical images, particularly focusing on 2D or 3D CT or MRI volumes. To enhance the efficiency of the annotation process, I have implemented an active learning method, reducing the workload for clinicians involved. My research primarily centers around the larynx region.To facilitate dataset annotation, I have integrated a Label Studio server, providing a robust framework. Additionally, I have successfully incorporated the interactive annotation capabilities of the Meta Segment Anything Model. Leveraging self-supervised techniques, I have effectively reduced the burden on doctors. Excitingly, my work with Transformers has yielded highly promising results, as the trained models can efficiently denoise medical images.

Conception and implementation of an augmented reality environment in which a technical partner can promote a new technology via the internet. Real-time video streaming, window frame show case instructions, and other tools are used to allow the user to make a remote teaching. This is a side-project which is provided by CERTH.

In order to approach AI-augmented decision support from surveillance footage, I am creating a Unity3D environment that simulates crowd behavior both indoors and outdoors. I am modeling a few major events such as fire, flood, and shoplifting, and then with the use of Deep Reinforcement Learning, I train an agent to perform appropriate actions based on the event.

I created an AR application as part of a larger project, using the MRTK framework. The HoloLens 2 application can handle UAVs via gestures or head poses, and the AR user can view drone video feeds. C# was widely used. Furthermore, using the FFmpeg library, I developed a C/C++ dynamic-link library Decoder for UWP devices (HoloLens 1/2) using as input raw h.264 video encoded frames and output a decoded frame ready for display. Performance and algorithm optimization was applied due to the low performance on HoloLens 1. Finally, using the DJI mobile SDK, an Android app has been developed to control DJI drones (DJI Mavic Mini, DJI Mavic 2 Enterprise Dual).

A full-stack position in a tourist area, with daily contact with new clients. Strong problem-solving capabilities and soft skills are commonly utilized.

A full-stack position in a tourist area, with daily contact with new clients. Marketing and SEO techniques are used to increase the possibility of gaining new clients. Strong problem-solving capabilities and soft skills are commonly used.

Administration of a LAN consisted of 50 workstations and 3 servers (Microsoft Windows Server 2008, 2012)Maintenance of all the IT and networking equipment and training of the personnel on new military applications.