AI Engineer | Applied Machine Learning & Language Models
I build practical AI systems, focusing on language-model-based applications and the engineering required to make them reliable—from model development and evaluation to deployment.
I'm particularly interested in efficient neural architectures and turning research ideas into working systems.
- TRLM — A ~75M-parameter language model built with shared-weight recursive Transformers for non-autoregressive generation. Accepted to the BabyLM Workshop at EMNLP 2026.
- S1GHT — A framework for real-time visual decisions (jev-like) on images and video, combining vision models, object tracking, and calibrated predictions.
- flycns-sim — An exploratory simulation project combining the Drosophila MaleCNS connectome, spiking neural models, and a FlyGym/MuJoCo physical body.
- EmbedTurk — Research on Turkish text-embedding models, published at INFUS 2025.
- MaskDriven-SD — A synthetic data generation pipeline using diffusion models and mask-based guidance.
- SerbestKategori-MedTurk — A Turkish multimodal medical assistant developed for the TEKNOFEST Turkish NLP competition.
Python · PyTorch · Hugging Face · vLLM · FastAPI · Docker
More projects and experiments are available in my repositories.