Quantitative modeller — thirteen years of mathematical modelling across domains, from continuum mechanics and constitutive modelling to financial markets. Currently completing an MSc in Data Analytics (Sabancı University, 2026).

I build models the same way regardless of domain: explicit assumptions, leakage-free protocols, calibrated probabilities, and an honest statement of what the data can and cannot support. The unifying thread is that models are not the deliverable — decisions are.

portfolio_projects — end-to-end projects, each running the full arc from EDA through rigorously tuned models to the decision layer and, where the data allows, causal analysis.

Current highlight: Bank Customer Churn as a decision system — six notebooks covering EDA, three tuned model families under an identical two-stage search protocol, probability calibration, a cost-matrix-derived campaign policy with sensitivity analysis, and a propensity-score causal analysis of the activation lever.

In progress: credit risk (PD / IFRS 9 language), classical time series (SARIMA/GARCH), deep learning forecasting (CNN + Transformer encoder), and an NLP/LLM track (recommendation, sentiment, RAG).

• Core: Python · pandas · NumPy · SciPy
• Machine learning: scikit-learn (pipelines, model selection) · imbalanced-learn · XGBoost · LightGBM · SHAP · statsmodels
• Deep learning: PyTorch — CNNs, BiLSTM + attention, Transformer encoders, self-attention implemented from scratch
• NLP / LLM: Hugging Face Transformers (incl. FinBERT) · gensim (Word2Vec) · sentence-transformers · TF-IDF pipelines · FAISS · RAG pipelines (chunking → embeddings → vector retrieval → generation) · local LLM deployment (Llama 3.1 8B via Ollama / Apple MLX)
• Data access: kagglehub · FRED & EVDS (CBRT) APIs
• Visualization & apps: matplotlib · seaborn · Plotly · Streamlit
• Tooling: Jupyter · VS Code · Git/GitHub · venv/conda · macOS (Apple Silicon)

LinkedIn — open to quantitative modelling and data science roles.