Ashot N.

Credit risk data scientist (building credit scoring models, scorecards, PD / loan default prediction and IFRS 9 ECL models in Python, ready for production and model review. I help lenders approve more good borrowers, cut losses, and make underwriting decisions they can stand behind. Models are built to go into production, not prototypes that get shelved. On a recent credit scoring project, my model lifted AUC from 0.81 to 0.85. In practice that meant the lender approved 8% more applicants while the default rate went down 1% — worth roughly 10% in added profit over a one-year horizon. Before Upwork, I spent 5 years as an in-house Data Scientist at one of Armenia's largest banks, building forecasting and risk models used in live operations. That experience shapes how I work: I care as much about how a model holds up over time and under review as I do about its accuracy on day one. I usually work with: → Fintech and alternative/SME lenders building or improving credit decisioning → Banks and credit unions that need scorecards built, rebuilt, or validated → Risk and analytics teams that need an extra pair of expert hands What I typically deliver: → Credit scoring and application/behavioral scorecards → Credit risk modeling: PD and loan default prediction models → Model validation, monitoring, and clear documentation → Risk segmentation and lending portfolio analytics → Fraud detection models for lending → Risk dashboards and SQL/Python pipelines that replace fragile Excel reporting → IFRS 9 / ECL support when a project calls for it Tools I rely on: Python (pandas, scikit-learn, statsmodels), XGBoost, LightGBM, logistic regression, and SQL, with FastAPI and Docker when something needs to be deployed. A few honest signals: Top Rated Plus, 100% Job Success, and $70K+ earned on Upwork, including work for US-based enterprise clients. One client's feedback on a recent credit scoring project was simply "Highly recommended," and another noted my "amazing knowledge in ML and DS." How I work: a short scoping call, then a clear plan with an honest estimate, then iterative delivery with documentation your team and any reviewer can actually follow. If something isn't the right fit for me, I'll tell you. Send a short message with your problem, your data, and your timeline. I usually reply within a few hours. The first 15-minute scoping call is free. Credit risk modeling · credit scoring · application & behavioral scorecards (WoE/IV) · probability of default (PD) · loan default prediction · IFRS 9 ECL · model validation · PSI/Gini/KS monitoring · risk segmentation · XGBoost · LightGBM · logistic regression · fraud detection · Python · SQL