Data Scientist · MLOps Engineer
I build the whole path — from the first notebook to a deployed, monitored API. Evaluated out-of-fold, so the number you see is the number you get.
The price model · 5-fold out-of-fold
End-to-end ML system on LightGBM with TF-IDF+SVD text features — scraping, dedup, leak-free 5-fold evaluation, drift monitoring, SHAP explainability, and a FastAPI serving layer. The complete production cycle, not just a notebook.
LightGBM · FastAPI · DuckDB · RailwayAn R package on CRAN for fuzzy meta-ensembles. It takes the validation predictions of several base learners — penalised regression, random forest, XGBoost, LightGBM — learns membership weights over that prediction space with Fuzzy C-Means, possibilistic FCM, Gustafson–Kessel or k-means, and fits one regression per cluster, tuned by grid search on validation loss. Nihat Tak is the author; I am a contributor (ctb).
R