I'm a Forward Deployed AI/ML Engineer with a background in Computer Science and Artificial Intelligence. I'm particularly interested in how models learn, how architectures can be improved and how research ideas can be turned into practical systems that people actually use.
Currently, I work as a Forward Deployed Founding Engineer.
A few areas I enjoy working on:
- Alternative neural computation paradigms
- Production-grade LLM systems and agentic AI
- Representation learning and model behavior
- Applied AI/ML systems built with a strong engineering focus
MSc Dissertation Project
Repository: github.com/berkedenizbozyigit/novel-artificial-neurons
Preprint: arXiv
My MSc dissertation explores one of the most basic assumptions in neural networks: that a neuron should aggregate its inputs through a weighted sum.
This project investigates alternative aggregation mechanisms and looks at how they affect learning dynamics, robustness, and representational capacity. As part of the work, I implemented F-Mean, Gaussian, and Hybrid neuron variants in PyTorch and evaluated them on vision benchmarks.
Selected Results
- 87.61% accuracy on CIFAR-10 vs. 87.33% baseline
- 0.991 robustness score vs. 0.890 for standard neurons
- Full implementation and experimentation in PyTorch
🥇 Departmental Prize Winner — awarded for the best project in the 2024/25 Artificial Intelligence MSc cohort at the University of York.
- Machine Learning and Deep Learning
- LLM-based Systems
- Neural Architectures and Representation Learning
- Explainable AI and Interpretability
- Retrieval-Augmented Generation (RAG)
- Agentic AI Applications
- Production AI/ML Systems
