Chemistry student at İstanbul Technical University who builds things with code.
I enjoy solving real-world problems with software — whether that's a data pipeline, a CLI tool, or a machine learning model. My chemistry background gives me a useful angle on scientific computing and drug discovery problems, but most of what I do lives in Python, Git, and the terminal.
- Software Development — Building practical tools and scripts in Python and C; comfortable on the command line, with Git, and on Linux/Unix systems.
- Data & Machine Learning — End-to-end ML pipelines with scikit-learn and XGBoost; feature engineering, model evaluation, and data wrangling with pandas and NumPy.
- Scientific Computing — Cheminformatics (RDKit), quantitative data analysis with Python & MATLAB, applying computational methods to chemistry problems.
Python C MATLAB scikit-learn XGBoost pandas NumPy RDKit Jupyter Git Linux
PPBR Prediction — A machine learning pipeline that predicts the plasma protein binding rate of drug compounds from their molecular structure. Combines Morgan fingerprints (2048-bit) with physicochemical descriptors (LogP, TPSA, HBD/HBA...) and a logit target transformation to handle PPBR's skewed distribution. Tuned XGBoost achieves R2 = 0.387 on a random split and R2 = 0.354 on a scaffold split — the more realistic test for generalisation to novel chemistry. Includes feature importance analysis, residual diagnostics, and cross-validation.
webscraper — A general-purpose CLI tool for extracting structured data from any website. Supports four extraction modes (emails, links, HTML tables, CSS selector), multi-page crawling with configurable pagination, and exports to Excel or CSV. Built with requests, BeautifulSoup4, and pandas.
gitpulse — A CLI tool that analyzes any Git repository and generates a visual terminal or HTML report. Covers commit history, top contributors, file churn, activity heatmaps, and language breakdown.
csv-detective — A CLI profiler for CSV files. Auto-detects column types, flags missing values and outliers via IQR, and summarizes distributions. Outputs a rich terminal report or a shareable HTML file.
admet-screen — A drug candidate screening pipeline combining rule-based filters (Lipinski, Veber, PAINS) with ML models trained on Morgan fingerprints. Predicts blood-brain barrier permeability and hERG cardiotoxicity risk, then scores each molecule with a composite ADMET score.
finpulse — A CLI portfolio analyzer that fetches live stock and crypto prices, calculates P&L and allocation, and uses Claude AI to generate a plain-language market summary with risks and opportunities. Exports an interactive HTML report with charts.