AI and research
models that analyse images, adapting language models, checking whether confidence scores are reliable, heat maps that explain image predictions
I design, train and publish AI models, then build the software needed to use them. My work includes models that choose among fixed answers and state their confidence, small models for AI assistants, and systems that detect AI-altered faces and show what affected each result.
I founded Eschatia Labs and work as an AI Engineer at Enocta. Before that I built financial software at KKB, backend and data systems at Alictus, and deep learning for smart farming at AgriDL5.0.
For my MSc at TED University, advised by Venera Adanova, I studied how varied and partly blacked-out training images affect a detector that examines video one frame at a time.
to present
AI Engineer
Deep learning, language models and AI engineering.
to
Full-stack Engineer
Financial software used across the Turkish banking sector.
to
AI and Full-stack Engineer (contract)
Deep learning for smart farming, a mobile app and work with government partners.
to
Backend and Data Engineer
Moved internal tools to the cloud using Go and Azure.
Internships and part-time roles: TED University (2023), SimBT (2019, 2021), Roar Games Studio (2021), BiSoft (2020).
Runs AI coding tools side by side from one local terminal workspace.
109GitHub7,379npm
Detects AI-altered face videos and explains predictions with facial-region heatmaps.
9GitHub784HF
A short list of contributions to external projects.
2023 to 2025
MSc in Computer Science
Full merit scholarship. Thesis on detecting AI-altered face videos and explaining model predictions, advised by Venera Adanova.
2022 to 2023
MSc in Computer Science and AI (one year, then transferred)
Full scholarship, remote.
2016 to 2021
BSc in Computer Engineering
models that analyse images, adapting language models, checking whether confidence scores are reliable, heat maps that explain image predictions
Python, Go, web and mobile development
communication between AI tools, running models locally
Turkish (native), English (C2, IELTS), German (B1, TELC)