I am a PhD candidate at NKUA, collaborating with NCSR Demokritos, with research interests in Neurosymbolic and Generative AI. I hold an MSc in Data Science and Machine Learning from NTUA and a BSc in Informatics and Telecommunications from NKUA. Currently, I work as a Principal ML engineer at ahedd - Digital Innovation Hub and recently stepped into a CTO role, leading teams for the first time. I have previously worked as a software engineer in the industry.
Developing and delivering core functionality features of our sportsbook platform.
Leading the development of a stress testing framework based on the NBomber framework to gather performance metrics for our distributed actor-model-based system using Grafana and Prometheus.
Being an active member of an Agile team, responsible for making product, software, and system design decisions.
Assisted in the design of the Public Key Infrastructure of various clients, based on nShield Connect HSMs and Microsoft PKI Services.
Lead Developer on the Mellon Remote Key Injection project, which is a Django-based web application responsible for the remote injection of cryptographic keys into POS devices.
Assisted in the development of the Mellon Receipts system. mReceipts is a Django-based transnational switch, integrating HSM-based security.
Part of the e-PIN Development Team, a .NET Framework MVC application, aiming at setting, retrieving, clearing, and verifying the banking pin of a customer in a safe cryptographic environment.
Lead the Development of the Mellon Instant Issuing / Foil Monitoring project. MICFMA is a .NET Core MVC application utilizing Javelin printing machines to facilitate the arising need for instant card issuing.
Assisted in the Development of the Mellon Pin File Management system. mPFM is .NET web application responsible for the credit card embossing process.
Software Engineer Intern:
Developed a .NET Windows Forms terminal (POS) logging system, utilizing SFTP so that the QA department was able to detect any possible issues, whilst complying with PCI requirements.
Developed a python TCP/IP library aimed at making HSM intercommunication, more developer-friendly. Became accustomed to with MyPy and MonkeyType.
Assisted in the development of a .NET Core intercommunication library.
Education
PhD in Knowledge Guided & Rule Validated Generative AI
MSc in Data Science and Machine Learning
BSc in Computer Science and Engineering
National and Kapodistrian University of Athens
Oct 2024 - Now
Generative AI models excel in tasks like image generation and language processing but often produce errors like hallucinations and artifacts due to over-reliance on patterns.
My research combines neural networks with symbolic AI to develop knowledge-guided, rule-validated systems.
By enhancing reasoning with methods like graph neural networks and automatic validation, it aims to create reliable outputs, with applications in fields like stable material generation guided by physical laws.
National Technical University of Athens
Sep 2022 - Sep 2024
GPA:
9.5 out of 10
I’m currently working on my thesis wherein we explore harnessing cross attention control for instruction-based auto-regressive audio editing.
Leveraging the Prompt-to-Prompt concept, originally used in image editing, we expand its functionality for intuitive audio manipulation. Users can edit audio content based on textual prompts without necessitating model modifications, retraining, or additional data.
This research pioneers the application of these techniques to auto-regressive text-to-audio models, showing promise for breakthroughs in AI-driven audio synthesis and creative expression.
Carried out my BSc thesis “Generating realistic nanorough surfaces via a Generative Adversarial Network” under the supervision of Dr. G.Giannakopoulos and Dr. V.Constantoudis.
We focused on how a Generative Adversarial Network (GAN) based approach, given a nanorough surface data set, can learn to produce statistically equivalent samples. Additionally, we examined how pairing our model with a set of nanorough similarity metrics, can improve the realisticity of the resulting nanorough surfaces. We showcased via multiple experiments that our framework is able to produce sufficiently realistic nanorough surfaces, in many cases indistinguishable from real data.
Worked on the paper Generating Realistic Nanorough Surfaces Using an N-Gram-Graph Augmented Deep Convolutional Generative Adversarial Network which is a derivative of my BSc thesis.
Presented our findings at 12th EETN Conference on Artificial Intelligence, SETN 2022.
Projects
PythonPyTorchTorchvisionNumpyPlotlySymPyray
RoughGAN
Generating Realistic Nanorough Surfaces Using an N-Gram-Graph Augmented Deep Convolutional Generative Adversarial Network.