Андрій

Andrii Korostashov
Junior ML Engineer / Data Scientist
Email: [open contact info](look above in the "contact info" section) | GitHub: github.com/andrii_github | LinkedIn: [open contact info](look above in the "contact info" section)
Location: Ukraine / Italy
Profile
Junior ML Engineer with a background in Computer Engineering and Applied Mathematics. Interested in machine learning, predictive modeling, time series forecasting, computer vision inference, and ML deployment. Experienced in building reproducible ML pipelines with Python, scikit-learn, statsmodels, TensorFlow/Keras, FastAPI, Docker, GitHub Actions, and model evaluation reports.
Technical Skills
Programming: Python, SQL, Git, Bash basics
Machine Learning: scikit-learn, XGBoost, TensorFlow/Keras, statsmodels, imbalanced-learn
Data Science: pandas, NumPy, matplotlib, seaborn, SHAP, time series analysis, classification, model evaluation
MLOps / Engineering: FastAPI, Docker, GitHub Actions CI, joblib, REST API, unit testing
Methods: Logistic Regression, Random Forest, Gradient Boosting, XGBoost, ARIMA/SARIMAX, SMOTE, GridSearchCV, threshold tuning, residual diagnostics
Projects
Customer Churn Classification
GitHub: github.com/AndryMilki/classification-models
Built an end-to-end customer churn prediction pipeline using Logistic Regression, Random Forest, Gradient Boosting, and XGBoost.
• Handled class imbalance with SMOTE and compared models against dummy, all-non-churn, and no-SMOTE baselines.
• Tuned hyperparameters with GridSearchCV and optimized the classification threshold on validation data.
• Evaluated final performance on a separate test set using F1-score and PR AUC; best model achieved F1 = 0.595 and PR AUC = 0.566.
• Added FastAPI prediction endpoint, Docker support, tests, CI workflow, SHAP/feature importance, and saved evaluation reports.
Industrial Production Forecasting with ARIMA/SARIMAX
GitHub: github.com/AndryMilki/forecast-ARIMA-model
Built a reproducible time series forecasting pipeline for the U.S. Industrial Production Index using ARIMA/SARIMAX models.
• Performed stationarity testing with the Augmented Dickey-Fuller test and analyzed ACF/PACF plots for model selection.
• Compared ARIMA/SARIMAX forecasts against naive and moving-average baselines using RMSE, MAE, MAPE, and Theil’s U2.
• Selected ARIMA(4,1,0) through information criteria and documented that simple baselines outperformed ARIMA on the selected test split.
• Added residual diagnostics, future forecasting, saved plots/tables, unit tests, CI workflow, and a reproducible project structure.
Website Defect Analyzer — CNN Inference Service
GitHub: github.com/AndryMilki/CNN_defector
Built a Dockerized ML inference service for website visual defect detection using CNN/Keras screenshot classification.
• Implemented FastAPI endpoints for URL analysis, image prediction, and health checks.
• Added Selenium-based screenshot capture and automated website quality checks.
• Containerized the service with Docker and docker-compose, added Makefile commands, environment configuration, and tests.
• Documented model limitations transparently: the project focuses on inference and deployment because the original training dataset is unavailable.
Education
V. N. Karazin Kharkiv National University
M.Sc. in Applied Mathematics 2025 – present
Double Degree Program with the University of L’Aquila, Italy
National Technical University “Kharkiv Polytechnic Institute”
B.Sc. in Computer Engineering 2021 - 2025
Languages
English - Upper-Intermediate
Ukrainian - Native
Russian - Native
Italian - Basic, if applicable