PoC-Gym: Towards More Reliable LLM-Assisted Proof-of-Concept Exploit Generation
Second Workshop on Large Language Models For Generative Software Engineering, 2026
Derin Gezgin
I am a third-year undergraduate student at Connecticut College, double majoring in Computer Science and Statistics and Data Science.
I am a member of two research labs at Connecticut College: the Autonomous Agent Learning Lab, where I work with Prof. Gary B. Parker and Jim O'Connor, and the Informatics Lab, where I work with Prof. Timothy James Becker. My research in the Autonomous Agent Learning Lab focuses on artificial intelligence for games and evolutionary robotics, while in the Informatics Lab I work on real-world applications of computer vision and machine learning.
From May 2025 to March 2026, I collaborated with Prof. Mayur Naik at the University of Pennsylvania and Prof. Saikat Dutta at Cornell University. I led the Docker integration for IRIS, containerizing the project workflow, publishing prebuilt images to Docker Hub for 189 Java projects, making them easier to build and run reproducibly. I also developed PoC-Gym, a system for LLM-assisted Java proof-of-concept exploit generation that combines CVE-tailored prompts, static source–sink traces, coverage-based feedback, and runtime validation.
I am currently working with Prof. Nick Cheney at the University of Vermont as part of the Neurobotics Lab through a Summer 2026 research internship, exploring continual learning and open-ended evolution.
Outside of research, I work as a teaching assistant and grader at Connecticut College. I am currently the Head Teaching Assistant of the Department of Computer Science at Connecticut College. I was also a member of the Connecticut College Computer Science Student Advisory Board as the Diversity Chair.
Publications
-
-
Evolutionary Optimization of Deep Learning Agents for Sparrow Mahjong
AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, 2025
-
Playing Atari Space Invaders with Sparse Cosine Optimized Policy Evolution
AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, 2025
-
A framework for river connectivity classification using temporal image processing and attention based neural networks
ACM Conference on Computing and Sustainable Societies, 2025
-
SCOPE for Hexapod Gait Generation
IJCCI Conference on Evolutionary Computation and Theory and Applications, 2025
-
Evolving Neural Controllers for Xpilot-AI Racing Using Neuroevolution of Augmenting Topologies
IJCCI Conference on Evolutionary Computation and Theory and Applications, 2025
Teaching Experience
I am fortunate to serve as a teaching assistant and grader at Connecticut College for the courses below, working with classes of 20–40 students. During the 2026–2027 academic year, I serve as the Head Teaching Assistant for the Department of Computer Science, coordinating a team of 25 teaching assistants. Previously, during the 2025–2026 academic year, I served as the COM110 TA Manager, coordinating the COM110 teaching assistants, session scheduling, and grading.
So far, as an undergraduate student, I have accumulated over 620 hours of TA service, supporting more than 200 unique students.
F: Fall Semester & S: Spring Semester & U: Summer Session
- COM110: Introduction to Computer Science and Problem Solving S24, F24, S25, U25, F25, S26
- COM212: Data Structures F24, S25, F25, S26, F26
- COM219: Computer Organization F25
- COM304: Algorithms F26
- COM316: Artificial Intelligence F26
- COM322: Computer Vision S26
- STA107: Introduction to Statistics S26
- STA207: Advanced Regression Techniques F25, F26