Computer Engineer
NLP · Automatic Speech Recognition · Turkic-language AI
Computer Engineering graduate from Marmara University. I build open, reproducible NLP and speech-recognition systems and datasets for Turkic languages — bridging academic research with engineering practice.
I work mainly on low-resource Turkic NLP and ASR — fine-tuning speech models, building morphological tools, and releasing open datasets, models and benchmarks. I also have hands-on experience in database administration and high-performance database architectures, and I develop end-to-end data solutions with Python and modern web tools.
E. Aydın, M. Kumcu, “TurkmenFST: A Comprehensive Rule-Based Morphological Analysis and Generation System for the Turkmen Language,”
Two-stage LoRA fine-tuning of Whisper Large V3 for general and medical Turkish; a 20-model ASR benchmark and an AcoSemantic evaluation. WER reduced 34.7% (relative) on an independent test set.
Rule-based analysis and generation for Turkmen, with the largest open-source Turkmen lexicon (32,000+ entries) served through a web interface. Published at IEEE USBEREIT 2026 (TurkLang 2026 symposium).
Multi-task detector on FakeAVCeleb fusing video (Xception), audio (Wav2Vec 2.0) and lip-sync streams via late fusion, with an ablation across modalities.
A two-branch digital-health idea: ECG analysis (arrhythmia detection) and an ECG training platform for medical students. Accepted into the Yıldız Kaşifleri (YTÜ) and Tech İstanbul NOVA (İBB) entrepreneurship programs.