[{"data":1,"prerenderedAt":175},["ShallowReactive",2],{"footer":3,"home-el":24,"publications":49,"portfolio-el":94},{"_path":4,"_dir":5,"_draft":6,"_partial":6,"_locale":7,"navigation":8,"text":5,"social":9,"_id":18,"_type":19,"title":20,"_source":21,"_file":22,"_stem":23,"_extension":19},"/footer","",false,"en","false",[10,14],{"name":11,"url":12,"icon":13},"Github","https://github.com/levgiorg","i-tabler-brand-github",{"name":15,"url":16,"icon":17},"Linkedin","https://www.linkedin.com/in/giorgos-levis/","i-tabler-brand-linkedin","content:footer.json","json","Footer","content","footer.json","footer",{"_path":25,"_dir":26,"_draft":6,"_partial":6,"_locale":27,"navigation":6,"intro":28,"name":29,"tagline":30,"description":31,"social":32,"contact":37,"resume":41,"_id":45,"_type":19,"title":46,"_source":21,"_file":47,"_stem":48,"_extension":19},"/sections/hero","sections","el","Γεια σας, είμαι ο"," George Levis","AI/ML Engineer","Φτιάχνω συστήματα που δουλεύουν, όχι παρουσιάσεις. Η καθημερινή μου δουλειά είναι μια enterprise πλατφόρμα agentic AI — LangGraph, MCP servers, autonomous LLM agents πάνω σε πραγματική data infrastructure. Παράλληλα βοηθώ επιχειρήσεις να βάλουν AI, αυτοματοποίηση και απλά καλό λογισμικό σε πραγματικά προβλήματα: διαδικασίες που τρώνε ώρες κάθε εβδομάδα, εργαλεία που δεν συνομιλούν μεταξύ τους, δουλειά που δεν θα έπρεπε να απαιτεί άνθρωπο. Αν βγαίνει σε παραγωγή και αντέχει, με ενδιαφέρει.",[33,35],{"name":34,"url":12,"icon":13},"GitHub",{"name":36,"url":16,"icon":17},"LinkedIn",{"button":38,"url":39,"icon":40},"Ας μιλήσουμε","mailto:levgiorg@gmail.com","i-tabler-message",{"button":42,"url":43,"icon":44},"Βιογραφικό","/GEORGE_LEVIS_CV.pdf","i-tabler-file-cv","content:el:sections:1.hero.json","Hero","el/sections/1.hero.json","el/sections/1.hero",{"_path":50,"_dir":26,"_draft":6,"_partial":6,"_locale":7,"navigation":51,"title":52,"description":55,"publications":56,"_id":91,"_type":19,"_source":21,"_file":92,"_stem":93,"_extension":19},"/sections/publications",{"title":52,"icon":53,"href":54},"Publications","i-tabler-book","#publications","My research publications in AI/ML and 6G Networks",[57,64,71,78,84],{"title":58,"authors":59,"description":60,"doi":61,"year":62,"publisher":63},"FedRA: A Fully Decentralized Federated Learning Framework for Robust Intelligence Sharing across O-RAN dApps","Sotirios T. Spantideas, Anastasios E. Giannopoulos, George A. Levis, Panagiotis Trakadas","We propose a resilient and fully decentralized federated learning framework specifically adapted for Open Radio Access Networks (O-RAN) named FedRA that enables collaborative intelligence between dApps. We present how FedRA can be deployed in the O-RAN architecture by using already existing interfaces between the well-defined RAN components and how real-time ML applications (dApps) can interact to share their distilled intelligence. The software components of FedRA are described along with the step-by-step deployment workflow, outlining practical guidelines for its implementation. Finally, the framework is validated in a simulated environment using both real and simulated datasets reporting the time series of throughput provided by multiple radio units; two different ML models hosted in FedRA nodes either detect anomalies in upcoming network traffic or forecast data-rate values. The achieved accuracy confirms the validity of FedRA in the training of cell-specific ML models and the decentralized collaborative intelligence sharing among them. FedRA is also compared to a typical client/server approach in terms of model accuracy and network communication cost.","https://doi.org/10.1109/MCOM.001.2500368","2026","IEEE Communications Magazine",{"title":65,"authors":66,"description":67,"doi":68,"techrxiv":69,"pdf":70},"A Peer-to-Peer Energy Management and Exchange Framework in Energy Communities via Actor-Critic Learning","George A. Levis, Sotirios T. Spantideas, Anastasios E. Giannopoulos, Member, IEEE, and Panagiotis Trakadas","The rapid growth of global energy consumption in buildings and the increasing adoption of renewable energy sources have created an urgent need for intelligent Home Energy Management Systems (HEMS). This work addresses the challenge of optimizing both individual home energy usage and community-wide energy distribution through peer-to-peer (P2P) trading. In the proposed framework, each smart home jointly optimizes the comfort maintenance, the energy cost and the community energy efficiency based on renewable generation, temperature dynamics, and grid prices. We propose a marketbased energy exchange mechanism with dynamic price setting and two complementary methods (threshold-based and rewardbased) to mitigate market manipulation in P2P energy trading. To enable decentralized coordination inside the energy community, the multi-agent Deep Reinforcement Learning (DRL) with actorcritic architecture is proposed to ensure the continuous and adaptive control of indoor temperature systems, battery storage, and price setting using real-world market prices and energy consumption data. The quantitative validation across multiple homes demonstrates a 39.3% reduction in grid dependency of the reward-based mechanism compared to unregulated trading. The framework maintains indoor temperatures consistently within comfort bounds while enabling intelligent battery management that adapts to price fluctuations, with state-of-charge levels peaking at 85% during low-price periods and decreasing to 40% during high-demand intervals. Overall, the approach establishes a self-sustaining energy ecosystem within residential communities, balancing market integrity and operational efficiency.","https://doi.org/10.36227/techrxiv.174802380.00970908/v1","https://www.techrxiv.org/users/923939/articles/1295878-a-peer-to-peer-energy-management-and-exchange-framework-in-energy-communities-via-actor-critic-learning","https://d197for5662m48.cloudfront.net/documents/publicationstatus/260000/preprint_pdf/8340e403d761df271f5d8e8ef97ee95f.pdf",{"title":72,"authors":73,"description":74,"doi":75,"year":76,"publisher":77},"Explainable Predictive Maintenance of Marine Engines Using a Hybrid BiLSTM-Attention-Kolmogorov Arnold Network","Alexandros S. Kalafatelis, Georgios Levis, Anastasios Giannopoulos, Nikolaos Tsoulakos and Panagiotis Trakadas","Predictive maintenance for marine engines requires forecasts that are both accurate and technically interpretable. This work introduces BEACON, a hybrid architecture that combines a bidirectional long short-term memory encoder with attention pooling, a Kolmogorov Arnold network and a lightweight multilayer perceptron for cylinder-level exhaust gas temperature forecasting, evaluated in both centralized and federated learning settings. On operational data from a bulk carrier, BEACON outperformed strong state-of-the-art baselines, achieving an RMSE of 0.5905, MAE of 0.4713 and R2 of approximately 0.95, while producing interpretable response curves and stable SHAP rankings across engine load regimes. A second contribution is the explicit evaluation of explanation stability in a federated learning setting, where BEACON maintained competitive accuracy and attained mean Spearman correlations above 0.8 between client-specific SHAP rankings, whereas baseline models exhibited substantially lower agreement. These results indicate that the proposed hybrid design provides an accurate and explanation-stable foundation for privacy-aware predictive maintenance of marine engines.","https://doi.org/10.3390/jmse14010032","2025","Journal of Marine Science and Engineering",{"title":79,"authors":80,"description":81,"doi":82,"acm":83},"SLEEPY-rApp: Delay-aware Sleep Scheduling for Energy Efficiency in MEC-enabled O-RAN","G. Levis, A. Giannopoulos, S. Spantideas, P. Trakadas","This paper presents a novel resource allocation and MEC server deactivation strategy for MEC-enabled O-RAN networks, with the objective of optimizing energy consumption while satisfying strict delay requirements. Our approach leverages an intelligent orchestration application, SLEEPY-rApp, deployed in the SMO layer to dynamically control MEC server activation and request routing. Simulation results indicate that our method significantly reduces energy usage while maintaining the required quality of service.","https://doi.org/10.1145/3721889.3721927","https://dl.acm.org/doi/abs/10.1145/3721889.3721927",{"title":85,"authors":86,"description":87,"doi":88,"arxiv":89,"pdf":90},"COMIX: Generalized Conflict Management in O-RAN xApps -- Architecture, Workflow, and a Power Control case","Anastasios Giannopoulos, Sotirios Spantideas, Levis George, Kalafatelis Alexandros, Panagiotis Trakadas","This paper introduces COMIX, a generalized Conflict Management scheme for Multi-Channel Power Control in O-RAN xApps. We propose a framework that detects and resolves conflicts between Deep Reinforcement Learning (DRL)-based xApps for power control, utilizing a Conflict Mitigation Framework (CMF) and Network Digital Twin (NDT). The research demonstrates significant network energy savings through intelligent conflict management in O-RAN environments.","https://doi.org/10.1109/ACCESS.2025.3585774","https://arxiv.org/abs/2501.14619","https://ieeexplore.ieee.org/iel8/6287639/6514899/11071295.pdf","content:sections:3.publications.json","sections/3.publications.json","sections/3.publications",{"_path":95,"_dir":26,"_draft":6,"_partial":6,"_locale":27,"navigation":96,"title":97,"description":100,"portfolio":101,"projects":102,"_id":172,"_type":19,"_source":21,"_file":173,"_stem":174,"_extension":19},"/sections/portfolio",{"title":97,"icon":98,"href":99},"Portfolio","i-tabler-clipboard-data","#portfolio","Εδώ θα βρείτε μερικά από τα έργα στα οποία έχω δουλέψει. Περισσότερα για τη δουλειά και τα ενδιαφέροντά μου στο",{"text":34,"url":12},[103,114,123,131,140,148,157,165],{"title":104,"description":105,"tags":106,"url":113},"DocuRAG","Σύστημα Retrieval-Augmented Generation (RAG) σε Python για ερωτοαπαντήσεις πάνω σε έγγραφα. Χτισμένο με FastAPI, HuggingFace transformers και FAISS vector search. Υποστηρίζει PDF, TXT, DOCX και MD, με REST API και web interface σε Streamlit.",[107,108,109,110,111,112],"Python","RAG","FastAPI","HuggingFace","FAISS","LLM","https://github.com/levgiorg/DocuRAG",{"title":115,"description":116,"tags":117,"url":122},"Smart Home P2P Energy Trading with RL","Προσέγγιση deep reinforcement learning για τη βελτιστοποίηση της ενεργειακής κατανάλωσης έξυπνων σπιτιών, με παράλληλη αποτροπή συμπεριφορών καρτέλ σε peer-to-peer αγορές ενέργειας. Υλοποιεί DDPG για τη λειτουργία HVAC, τη διαχείριση μπαταρίας και τη στρατηγική τιμολόγησης.",[107,118,119,120,121],"PyTorch","DDPG","Multi-Agent RL","Energy Trading","https://github.com/levgiorg/Smart-Home-P2P-Energy-Trading-RL",{"title":124,"description":125,"tags":126,"url":130},"Finetuning LLM Gemma3 for AI Detection","Σύστημα ανίχνευσης κειμένου παραγόμενου από AI, μέσω fine-tuning του Gemma3-4B της Google με βελτιστοποιήσεις Unsloth στο dataset RAID. Ανιχνεύει κείμενο από 11 διαφορετικά LLMs σε πολλαπλά πεδία, με εκπαίδευση αποδοτική σε μνήμη πάνω σε καταναλωτικό hardware.",[107,112,127,128,129],"Fine-tuning","Gemma","NLP","https://github.com/levgiorg/Finetuning-LLM-Gemma3-AI-Detection",{"title":132,"description":133,"tags":134,"url":139},"MNIST MLOps Pipeline","Pipeline ML έτοιμο για παραγωγή με Kubeflow Pipelines, MLflow και KServe. Καλύπτει ολόκληρη τη ροή MLOps, από τη συλλογή δεδομένων έως το serving του μοντέλου, με παρακολούθηση πειραμάτων, containerized components και infrastructure as code.",[107,135,136,137,138],"Kubeflow","MLflow","KServe","MLOps","https://github.com/levgiorg/Mnist-example-in-kubeflow",{"title":141,"description":142,"tags":143,"url":147},"TARDIS-Pruning","Ελαφριά βιβλιοθήκη pruning νευρωνικών δικτύων για μοντέλα PyTorch, χτισμένη πάνω στο Microsoft NNI. Υλοποιεί L1 norm pruning για μείωση του μεγέθους και των υπολογιστικών απαιτήσεων σε edge deployment, διατηρώντας την απόδοση.",[107,118,144,145,146],"Model Compression","NNI","Edge AI","https://github.com/levgiorg/Pruning",{"title":149,"description":150,"tags":151,"url":156},"Fedra: Federated Learning Framework","Συγκριτική μελέτη αποκεντρωμένων και κεντρικοποιημένων αρχιτεκτονικών federated learning. Επεκτείνει το P2P framework Fedra με σύγκριση έναντι του Flower, ιεραρχικό FL και πειράματα σε συνθετικά και πραγματικά δεδομένα, μεταξύ άλλων πρόβλεψη κυκλοφορίας με LSTM.",[107,152,153,154,155],"Federated Learning","P2P","LibP2P","Flower","https://github.com/levgiorg/fedra",{"title":158,"description":159,"tags":160,"url":164},"SnakeAI-DQN","Υλοποίηση Deep Q-Network (DQN) για την εκπαίδευση agent που παίζει το κλασικό Snake με reinforcement learning. PyTorch για το νευρωνικό δίκτυο, Gym για το περιβάλλον και Pygame για την απεικόνιση, με πλήρη οπτικοποίηση της εκπαίδευσης.",[107,118,161,162,163],"DQN","Reinforcement Learning","Gym","https://github.com/levgiorg/SnakeAI-DQN",{"title":166,"description":167,"tags":168},"Multi-Agent Energy Management System","Framework προσομοίωσης κατανεμημένης διαχείρισης ενέργειας που μοντελοποιεί πολλαπλά οικιακά συστήματα HVAC και μονάδες αποθήκευσης. Αλγόριθμοι reinforcement learning βελτιστοποιούν την κατανομή ενέργειας στο δίκτυο, τηρώντας τους περιορισμούς θερμοκρασίας.",[107,118,169,170,171],"DRL","Multi-Agent Systems","Energy Management","content:el:sections:5.portfolio.json","el/sections/5.portfolio.json","el/sections/5.portfolio",1786519415126]