Nastija Kalugina

On the RSNA Knee Abnormality Detection Kaggle competition, I built a knee MRI pipeline from DICOM images to 2.5D models using PyTorch and timm. I also extracted training labels from radiology reports with an ensemble of LLMs and used patient-level five-fold validation; my public leaderboard score was 0.949.

For product analytics, I answered 11 business questions about the Olist marketplace in SQL and developed an A/B-test workflow for Criteo Uplift. It included power analysis, SRM and balance checks, bootstrap, CUPED, and Holm correction.

I also created a four-quarter forklift market forecast with prediction intervals, backtesting, and interest-rate scenarios. For a customer churn project, I served a scikit-learn model through FastAPI, Docker, pytest, and GitHub Actions; its ROC-AUC was 0.867. I'm studying Data Science at Нетология and looking for a remote junior role.