Data Scientist / ML Engineer
I'm a 2026 graduate of Astana IT University (Big Data Analysis) looking for a junior Data Scientist or ML Engineer role in Kazakhstan.
During my internship at KTZ I built ticket-sales forecasting models on ~6M records with CatBoost and LightGBM, and for my thesis I developed a career recommendation system. My main portfolio project is Home Credit Default Risk (ROC-AUC ~0.79), which I wrapped into a FastAPI, PostgreSQL and Docker service. Right now I'm building an xG (expected goals) model on Wyscout football data with time-based validation. I enjoy running experiments and watching the metrics improve.
Kazakhstan Temir Zholy
Mar 2026 — Jun 2026Built ticket-sales forecasting models (3/10/30-day horizons) on ~6M railway records over ~40 months using CatBoost and LightGBM, with Optuna tuning, MLflow tracking and SHAP analysis. Improved forecast error by ~21–22% over a moving-average baseline.
Astana IT University
2023 — 2026Thesis (defended June 2026): career recommendation system for IT students: calibrated CatBoost classifier, TF-IDF skill matching, LightGBM labor-demand forecasting, course retrieval and an LLM advisor, deployed with FastAPI, React and Docker Compose.
Loan default prediction on the multi-table Home Credit dataset. Built feature engineering across the related tables, trained and tuned CatBoost, LightGBM and XGBoost with Optuna, validated with 5-fold stratified CV, out-of-fold predictions. Result: ROC-AUC ~0.79 (Kaggle private score 0.791). Wrapped the model in a FastAPI and PostgreSQL service that returns the default probability for a client ID.
Open projectBuilt CareerFlow, a deployed ML career guidance app for IT students. Combined CatBoost classification, TF-IDF skill matching and LightGBM historical demand scoring to rank 7 career paths. Added personalized roadmaps and Claude/Gemini AI chat. Built with FastAPI, React, PostgreSQL and Docker; deployed on Azure. https://careerflow.live
Open projectShot-level expected-goals (xG) model on the open Wyscout dataset (league matches 2017–18, Euro 2016, World Cup 2018). Strict chronological validation: expanding-window CV over time blocks, with May 2018 and the World Cup held out as the test set. CatBoost baseline: CV log loss 0.288, ROC-AUC 0.789, mean predicted xG 0.110 vs actual goal rate 0.112. Work in progress: adding player and team form and
