Batu K. Masak
Undergraduate researcher at Özyeğin University, double-majoring in AI & Data Engineering and Computer Science. My work focuses on adversarial robustness of large language models, NLP, and the security implications of multi-turn interactions with AI systems.
GitHub profile: github.com/batukoray
Research Interests
Adversarial Attacks on LLMs
Multi-turn jailbreaking strategies that exploit rhetorical persuasion rather than gradient-based optimization. Bridging social engineering and machine learning safety.
NLP & Transformer Architectures
Language modeling, attention mechanisms, and the computational foundations of modern NLP systems. Particular interest in Turkish morphological complexity.
AI Safety & Alignment
Red-teaming methodologies, robustness evaluation, and understanding how safety fine-tuning can be systematically circumvented.
Mechanistic Interpretability
Understanding the internal representations and decision pathways of neural networks, with a focus on how models encode and process adversarial inputs.
Projects
Korado
↗CLI linear programming and mixed-integer programming solver built on PuLP/CBC backend.
HaDi — Hacker Dispatch
↗Curated cybersecurity and hacking intelligence dispatch — aggregating vulnerability disclosures, exploit research, and security news.
LLaMA From Scratch
↗Ground-up implementation of the LLaMA architecture for CS449 NLP course project at Özyeğin University.
El Pulpo
↗Personal terminal productivity system with MCP integration for task management, notes, and intelligent workflows.
Education
B.Sc. AI & Data Engineering · B.Sc. Computer Science
Double major · GPA: 3.72/4.00 · Ranked 1st in department (out of 54)