M.Sc. Data Science student at Dokuz Eylül University, working on both sides of applied AI: LLM systems (agents, RAG, evaluation) and classic machine learning (risk scoring and anomaly detection on imbalanced data). My thesis is on the statistical evaluation of open-source LLMs with LLM-as-a-Judge. I like building things end to end: the model, the API, and the tests that show it works.
- 🎓 M.Sc. in Data Science @ Dokuz Eylül University (in progress, GPA 3.70) · B.Sc. in Electrical & Electronics Engineering
- 🔬 Thesis: statistical evaluation of 7B–8B open-source LLMs (Friedman, Wilcoxon, McNemar)
- 💼 AI Model Evaluator (RLHF) @ Outlier AI, part-time
- 🧪 AI & Computer Vision Engineer @ Glossa, an early-stage startup project (12/2025 – 07/2026) · 1st place, LLM Jetpack Ideathon
- ⚡ Former project engineer in solar energy
- 🌍 Top 10, EU-funded E-DATA Project (study tour in the Netherlands & Germany)
- 📫 Open to junior AI Engineer and Data Scientist roles (EU or remote) · EU citizen
Also regularly: XGBoost · SHAP · SQL · PySpark · LangGraph · Groq / Llama-3 · MLflow · Pinecone
| Project | What it is | Stack |
|---|---|---|
| Electricity Theft Detection | Risk scoring on meter data of 42K customers to prioritize field inspections. The top 20% riskiest customers cover 63% of theft cases. Includes SHAP explanations and a Power BI dashboard | XGBoost, SHAP, Power BI |
| Explainable Credit Risk Scoring | Default-risk model with credit metrics (KS/Gini), SHAP explanations, and LLM-generated Turkish rationales | scikit-learn, XGBoost, SHAP |
| Responsible AI Banking Agent | A LangGraph banking assistant, hardened for regulated use: PII redaction (TCKN/IBAN), guardrails, audit logging, model card | LangGraph, Presidio, Groq |
| Agent Benchmarking Lab | Compares ReAct vs Plan-and-Execute agents with a statistical layer and BigQuery logging | BigQuery, SciPy, Streamlit |
| AI Support Pipeline | Support tickets auto-categorized and prioritized by an LLM, served over a FastAPI API with a |