resume
download pdfai engineer and grad student. i work on refusal and safety in llms through interpretability, and at SCL on a new way of doing market research.
boğaziçi
physics, on a hill over the bosphorus. loved the research and the inspiration from my profs, one of the best decisions i made in my life. in the EarthML research group Arkadaş hoca taught us machine learning and deep learning, and i worked on earthquake detection models and extensive data analysis. i also taught numerical methods, and once we even turned seismic data into an art exhibition. then i continued down the data science path, with internships at Allianz and Live The World.
the habit
physics left me one habit: when something works, i want to know what is actually happening inside it. first it was physics, then earthquakes, now language models. same question.
padova
a computer science master's in padova. i didn't quite like the program, or let's say i didn't fit in. after the first year of coursework i didn't want to do any research there, maybe i couldn't meet the right people at that time. the long version is on the blog, in turkish.
sabancı
back in istanbul, a data science msc at sabancı. working with Dilara Keküllüoğlu, i had my research spark again, with refusal dynamics and poisoning attacks on llms. my thesis is on the same topic with more interpretability candies on top: i look at what's happening while a model refuses a harmful request and try to make sense of the math inside it. i am also a ta for quantum programming on the side.
back in istanbul i also met Berkan Cesur, who taught me a lot about product development. can't lie, amazing person and insane inspiration. in one year with him i got experience in web development, app development, product cycles and a lot more.
now
days are SCL. i started as an ai engineer, and with time i took on more responsibility, so i could say i am a bit more than that now. Ömer Ülgen, our ceo, is one of the sharpest people i have worked with, and from Sinan Ülgen i am learning a lot about customer relationships and working B2B. nights are still the thesis.
AI Engineer · Synthetic Consumer Lab
Engineering lead responsible for the full product stack: backend, frontend, AI systems, and statistical methodology.
+ detail− detail
- Architected the platform on Laravel, Python/FastAPI, and Redis
- Built agentic systems for market research, and synthetic consumer personas grounded in real demographic and behavioral data
- Built agnus, an AI agent for market research
Teaching Assistant · Sabancı University
Teaching assistant for Quantum Programming, guiding students through quantum computing concepts, circuit design, and hands-on implementation with quantum programming frameworks.
AI Engineer / Data Scientist · SoundBoost
Designed and deployed deep learning models for audio source separation, classification, and acoustic event detection.
+ detail− detail
- Built end-to-end AI pipelines, with Django backends for model serving and JavaScript for real-time inference
- Led development of AI agents for complex audio processing workflows
- Built SoundBoost's free tools, including Loudness Penalty and a LUFS meter
AI Engineer Intern · Live The World
Built content generation pipelines using LLMs and web scraping.
+ detail− detail
- Improved the web scraping stack and developed Python tools for AI-driven applications
Data Analytics & Process Mining Intern · Allianz TR
Automated Excel reporting workflows with Python and SQL, built dashboards for operational visibility, and optimized business processes with Celonis process mining.
Research Assistant · Boğaziçi University
Worked on feature engineering and transformer-based architectures for seismic data analysis and earthquake detection.
+ detail− detail
- Contributed to the paper arXiv:2407.18402
Teaching Assistant · Boğaziçi University
Led problem sessions for Numerical Methods, teaching NumPy, SciPy, and Matplotlib through hands-on problem solving.
M.Sc. in Data Science · Sabancı University
Thesis on the mechanistic interpretability of large language models: how refusal and related concepts are represented geometrically in a model's internal activations. Coursework in advanced deep learning and statistical analysis.
Graduate Studies in Computer Science · University of Padua
Completed the first year of the M.Sc. program, with coursework in artificial intelligence, deep learning architectures, and algorithmic problem solving.
B.Sc. in Physics · Boğaziçi University
Member of the EarthML research group and the Science Club.
TÜBİTAK 2209-A
Developed an earthquake detection model, with most of the work in feature engineering.
Earth-ML
Time series classification for geophysical data.
Kaggle ML Challenge
8th place in the Türkiye İş Bankası ML Challenge 5, mainly through feature engineering.
Datathon AI
3rd place in a computer vision competition.
NLP News Summarization
Comparative evaluation of BART and T5 for summarization.
"Burası" Art Exhibition
Turned seismic data into artwork for an exhibition.