- Developed Airflow and Kubeflow ML pipelines which included data collection,model training / prediction, saving scores to database and metrics calculation- Configured pipeline failure alerting to Slack, set model dashboards in Grafana- Developed several REST API services to obtain model predictions in real time- Worked on a customer support chatbot powered by GPT technologies- Solved ad-hoc tasks using SQL queries to Clickhouse, Vertica and PostgresAchievements:- Created auxiliary Python library for quick pipeline deployment to Kubeflowthat reduced average deployment time by 50%- Received top-10% performance rating in Q4 2022- Received "Consistent star" badge based on 9-box grid talent evaluation in 2023 (top-5%)

After being promoted to a team lead, my main tasks were:- leading the validation of models developed in three departments of Sberbank- communication with departments' chief data scientists: scheduling model validation, negotiation of methodology, planning joint projects- managing a team of five data scientists and taking full responsibility for their work- creation of a validation methodology for new types of developed models (CV, clustering)- implementation of planned priority projects (quarterly/annual department goals)- preparation of different progress reports for the department director.Achievements:- optimized several validation tests, reducing their computation times by 10-100 times without a significant quality loss- conducted a pilot of the SberbankDS platform for colleagues from Kazakhstan and Belarus- since the beginning of my work at Sberbank, took part in the training of about 15 people and interviewed 80+ candidates.

Was responsible for the validation of various machine learning models, such as binary/multiclass classification, regression, time series, and text mining.Validation is a comprehensive evaluation of a machine learning model that includes qualitative and quantitative analysis as well as alternative modeling on the original data samples.Key achievements:- developed and continually improved a Python library which performs all validation tests in sequence and generates an Excel report for classification and regression models. To date, the library is actively used: 300+ models have been tested using it- participated in the development of internal data science platform "SberbankDS" and integrated all validation tests from the local library to the platform- created a Python module to automate routine actions in internal websites using requests library (i.e. changing statuses of 180+ tasks in the model tracking system, recreating 50+ test nodes on the DS platform)- conducted a course on data analysis for participants of the Sberbank MBA program (got a teacher's score of 9.1 out of 10 based on feedback).

Developed Python modules called "nodes" to be run on the DataFlow platform according to customer needs. Examples of nodes are "Read CSV", "Row Filter", "Fit Random Forest", "Calculate metrics", "Send SMS", etc. The project is built from these nodes and can be launched or scheduled to run at a certain time.It was a freelance job that I did in my spare time.Other tasks included:- integration with databases to save and load data- setting up and maintaining real-time projects using Kafka- building docker images with sufficient environment to run nodes- conducting demonstrations of the platform for potential customers.

Led a team of three data scientists in a CRM Department:- built client behavior models (such as response probability on a product offer) in a KNIME environment using machine learning methods- wrote R scripts to process data, calculate attributes, select features, and fit models- created an infrastructure for a regular model rebuild, including feature selection, hyperparameters tuning, model fit, client base scoring- developed an analytical data mart containing all kinds of client indicators: demographics, balances, transactions, turnover dynamics, open products, etc.- made several back-tests to analyze the impact on Bank's sales after ML models implementation- constructed a regularly-updated report on the quality of models, including key variables, ROC AUC, Lift, true/predicted values comparison.

Worked in a Data Analysis Unit of a Retail Business Department:- issued and reworked regular financial reports in Excel; forecasted the execution percent of quarterly business plan- developed a consolidated report on the areas of retail business, created a structure for automatic data update in presentation for a Head of retail department- wrote ad-hoc SQL queries and stored procedures on Teradata, calculated different attributes for financial models- optimized SQL-queries taking into account the specifics of the Teradata architecture (primary indexes, data partitioning, statistics collection, a mechanism for prioritizing queries through workloads).

Developed data transformation processes (ETL) running in a Corporate Data Warehouse using SAS Data Integration Studio:- configured and debugged data flows in SAS Management Console- wrote and optimized SQL-queries to Teradata Database using system tables- used Bash scripts to work with the file system and regular expressions to parse text data- communicated with colleagues on the customer side.