Iuliia Krivorot

- Co-developed a multi-agent simulation system to model global economic scenarios and commodity-market dynamics using agent-based modeling (LangGraph). - Designed and deployed ML models (MAPE < 10%) forecasting oil, non-ferrous and precious metal prices; production pipelines integrated via MLflow and Docker - Established monitoring dashboards in Power BI and Grafana to track model drift and performance metrics (MAPE, RMSE), reducing unplanned model retraining by 30%.

- Validated 15+ production models (OLS; XGBoost; ResNet, EfficientNet; BERT, RoBERTa) using statistical tests and cross-validation (statsmodels, scikit-learn). - Automated the validation pipeline (Python, pytest, GitLab CI), cutting testing time by 25% - Pioneered department’s first generative model (GigaChat, Kandinsky) validations for risk and compliance - Leveraged Optuna and Hugging Face for tuning, boosting accuracy 3–5% with no added latency - Authored 15+ detailed validation reports (Word, Excel)

- Built a collaborative filtering recommendation engine from scratch for a tax advisory web portal; increased user engagement metrics by 18% within three months of launch. - Developed interactive Power BI dashboards visualizing key business KPIs, reducing monthly report generation time by 50%. - Executed R&D on satellite-to-map image matching: manually labeled 1,000+ images; benchmarked Vision Transformer and CNN architectures (OpenCV, PyTorch) to achieve 92% geographic tile classification accuracy.