PhD researcher in AI (Computer Engineering, Bahçeşehir University). I build applied AI systems — LLMs, retrieval-augmented generation, speech — and study the evaluation methods that tell us whether they actually work.
- Agentic and graph-based retrieval-augmented generation
- LLM evaluation and hallucination detection
- Human verification and oversight in agent-driven workflows
- Responsible release of sensitive NLP datasets
- AI for education and assessment
- speechgauge — offline pronunciation, fluency, and prosody scoring (GOP, semitone prosody, forced alignment). No cloud. · write-up
- co-draft — human-in-the-loop LLM text editing: the model proposes span edits, you approve them as visual diffs. · write-up
- PrivHSD-Challenge — privacy-aware workflow for releasing and reviewing harmful-speech datasets. Built at the Council of Europe Democracy Hackathon.
- Speaklify — ASR + LLM pipeline for CEFR-aligned speaking assessment and feedback.
- turkish-math-curriculum-nlp — unsupervised NLP analysis of curriculum change in Turkish mathematics education.
