projects
mostly the thesis these days: where refusal lives inside llms. the older ml projects moved down to earlier work, they had their time.
refusal geometry in llms
msc thesis at sabanci: how refusal and harmfulness live in the internal geometry of llms. ask a model something harmful and it refuses, that refusal can be shown as a direction in activation space, and jailbreaks work by pushing the model off it. i am mapping what those attacks actually do to the representations, no public repo yet, we will see where it goes.
earlier work
Neural Text Summarization: Comparative Analysis of Transformer Architectures
Comparative study of BART and T5 architectures for automated news summarization, including sentiment-aware evaluation and ROUGE-based benchmarking.
Multi-Label Classification System for Financial Recommendations
Built a multi-label recommendation system for Isbank using one-vs-all XGBoost with feature engineering and ensemble strategies for improved predictive performance.
Machine Learning Algorithm Implementations
Collection of practical machine learning implementations, including boosting, ensemble techniques, and feature engineering workflows applied to real-world datasets.