Currently AI Research Engineer · MSc Student @ Istanbul Technical University My interest is building research-oriented ML systems. I am currently developing a multidisciplinary metallurgical microstructure analysis platform.

TANGEM — Graph Generation via Temporal-Aware Biased Walks · TMLR 2026 A Transformer-based graph generator that injects a temporal-similarity bias into random-walk sampling and models the resulting sequences autoregressively with a decoder-only Transformer. Benchmarked against 7 generative baselines (including DiGress and GraphRNN) across 4 datasets, it achieved the strongest structural fidelity on large & sparse networks under single-input case while scaling in linear time. PyTorch · Transformers · Graph ML

Movie Recommender Agent A conversational movie recommendation agent that pairs collaborative-filtering signals with an LLM, exposed through an MCP tool server and a FastAPI backend. A hands-on exploration of LLM tooling — wiring recommendation logic into an agent that can be called as a tool. Python · FastAPI · MCP · LLM

IBB Traffic Graph Data — Benchmarking & Prediction · IEEE CAMAD 2024 A road-traffic congestion prediction model on a novel 2,451-sensor Istanbul traffic dataset, combining GLEE graph node embeddings with engineered temporal features and an ExtraTrees classifier. Benchmarked across 4 embedding methods and 7 classifiers, it reached 91.3% F1 score, a 4% average gain over baselines. The repo also releases the benchmark dataset. Graph Embeddings · scikit-learn · Time Series

One-Shot Signature Identification · IEEE SIU 2025 A one-shot signature identification pipeline that fine-tunes an ImageNet-pretrained ResNet50 with Supervised Contrastive (SupCon) loss, trained on a synthetic dataset derived from Omniglot. It reached 91% one-shot accuracy on 200 unseen signatures, a 20% gain over the non-contrastive baseline. PyTorch · Contrastive Learning · Computer Vision

LinkedIn · Email