Пожалуйста, войдите, чтобы скачать резюме.
Experience level:
1-3 years
Education level:
Bachelor
Languages:
Russian: Native
English: B1
Calibration of the PD model
He regularly calibrated the model to adapt to the current portfolio.
Increase the accuracy of forecasting the probability of default and reduce the bank’s risks.
Improved the stability of the model by increasing the calibration to ±2% of the actual default level.
Development of a new version of the PD model based on new sources
Integrated data from the BCI and internal scoring.
The expansion of the feature space and the growth of predictive power.
Gini increased by 7 percentage points, which increased the effectiveness of scoring when issuing loans.
Collaboration with data engineers on the development of the Feature Store
Participated in the design and coordination of the storage structure of the features.
Unify data access for different models and speed up experiments.
The time for preparing datasets for training models has been reduced from several days to hours.
Writing documentation on the application funnel and products
He recorded agreements on model versions, update dates, and business restrictions.
For transparency of interaction within the team and preservation of expertise.
The documentation reduced the onboarding time for new team members by ~40.3%.
Creating and maintaining dashboards by metrics (Gini, UAT, PSI)
He has developed interactive dashboards in BI-tools.
To monitor the quality of models and operational degradation control.
The time for metric analysis was reduced by 3 times, and it was possible to quickly identify model degradation (PSI > 0.2).
Automation of the data processing pipeline: A reliable data analyzer has been developed for processing and standardizing various data sets provided by external partners, which ensures stable data quality for subsequent modeling.
Function development and selection: Comprehensive research data analysis (EDA) was conducted to identify key risk factors. Methods of selecting characteristics have been introduced to refine the functional space with an emphasis on financial stability and liquidity indicators.
Dataset optimization: The data set generation process has been simplified, manual intervention has been reduced, and the transition from providing raw data to modeling-ready functions has been significantly accelerated.
python, numpy, pytorch, tensorflow, postgresql, scikit-learn, mysql, seaborn, sql, matplotlib, jupyter notebook, scipy, hadoop, prometheus, docker, fastapi, catboost, computer vision, tableau, deep learning, clickhouse, big data, mongodb, kubernetes, анализ данных, api, xgboost, ci/cd, gitlab, pandas