Dmitry Svirchkov

• Developed machine learning model to predict potential cheaters before first trade and place them under surveillance monitoring.
• Developed machine learning model to detect and alert ongoing suspicious activities, providing timely notifications to anti-fraud monitoring specialists.
• Developed approach of merging several anti-fraud models to single score.
• Researched historical data with causal inference methods to estimate clients' sensitivity to various treatments / policies, when real AB-testing was not possible.
• Made several deep customer researches in fields of marketing, trading and anti-fraud.

Picked up the local credit scoring project for installment cards and transformed into a full-fledged ML production scoring service for external b2b-clients.

• Full technical ownership of project including model development, data analysis, MLOps, sales support.
• Made a research of explaining complex black box models using SHAP, PDP & ICE plots, shallow decision trees, linear models.
• Researched which "hard-to-classify samples" can be discarded from the model, in order to improve quality of remaining objects
• Added new data sources to models, which increased coverage and accuracy
• Developed model stability monitoring to prevent data-drift.

Customer research and analytics.

• Developed a churn prediction model to forecast account balances of inactive clients
• Made time series forecasts for budgeting purposes.
• Analysis exploring potential outcomes of converting a free debit card into a paid one.
• Developed multiple dashboards in Tableau

Developed several data-products tailored to upsell to b2b-clients.

• Campaign analytics with multi-channel attribution model based on Markov chains.
• Audience segmentation powered by unsupervised machine learning techniques.
• Customer behaviour analytics for clients' website.