Computer Engineer · Data Scientist
I build machine-learning systems — deep learning, reinforcement learning and graph ML — from research to deployment.
Open to work, 2026 / MSc Data Science, Sapienza / Roma
fig.01 — domain graph · hover a node to propagate
Computer Engineer and Data Science MSc candidate at Sapienza University of Rome. Before Roma, Computer Engineering at OSTIM Technical University (GPA 3.04).
I work across the machine-learning stack — deep learning for vision, reinforcement-learning agents, graph neural networks, and end-to-end data-science pipelines — and like taking models from research all the way to deployment.
My MSc thesis applies graph neural networks to anomaly detection in water-infrastructure networks. Backed by R&D internships in embedded and government-scale systems. Most interested in reinforcement learning, deep learning and MLOps.
Sapienza, Roma · Thesis
Adversarial self-play GNNs for water-distribution network cyber-defence (Supervisor: Prof. Francesca Cuomo). A GraphSAGE + GRU + Mixture-of-Experts defender reconstructs missing pressure/flow readings and flags compromised sensors; a Stackelberg attacker↔defender loop generalises to attacks unseen in training. On the Modena network: anomaly F1 0.767 (+5.8%), pressure error −23%, reproducible across seeds. Python · PyTorch Geometric · GraphSAGE · GRU
Sapienza, Roma · Research
Disease-gene prediction and drug repurposing for anxiety disorders. DIAMOnD and heat-diffusion methods over STRING PPI interactomes ranked candidate risk genes and repurposing hits — Buspirone surfaced as the top candidate. Python · NetworkX · Network Analysis
Personal project · Self
Real-time, emotion-aware music recommender. Reads user emotion from webcam input and plays context-aware playlists through the Spotify API, behind an adaptive UI. Python · OpenCV · Deep Learning · Spotify API
KAREL Electronics · Ankara
Embedded-systems R&D in C/C++ on Linux, plus a responsive internal system-monitoring website. C/C++ · Linux · Embedded · HTML/CSS
Revenue Administration · Ankara
Backend and data engineering on government-scale information systems and a CV-filtering web app. Wrote secure backend data-handling logic. Java · Spring Boot · PostgreSQL · Docker
OSTIM Tech University · Leadership
Ran technical webinars, networking events and hackathons for the OSTIMTECH student branch. Leadership · Community
MSc thesis. A GraphSAGE + GRU + Mixture-of-Experts defender reconstructs missing pressure/flow readings and flags compromised sensors in water-distribution networks. A Stackelberg attacker↔defender self-play loop generalises to attacks unseen in training. Modena network: anomaly F1 0.767 (+5.8%), pressure error −23%.
Disease-gene prediction and drug repurposing: PPI interactomes, DIAMOnD + heat-diffusion gene discovery, enrichment analysis, and repurposing candidates ranked over STRING.
Real-time emotion CNN trained on FER2013, deployed via TFLite, run live through OpenCV.
Reads emotion from the webcam and plays matching playlists via the Spotify API, with adaptive RL behaviour.
From-scratch DQN, A2C, DDPG and PPO on CartPole-v1.
Three engines — keyword, TF-IDF/cosine, and a custom score — over scraped degree data.
From-scratch MinHash recommender over Netflix-like data, with clustering.
Centrality, shortest paths, graph disconnection and community detection.
End-to-end pipeline: cleaning, feature engineering, selection and modelling.
Full index at github.com/onurozansunger ↗
Email is the fastest way. Open to research, full-time roles, or a quick chat.
onurozansunger@gmail.com