Evgeny Patekha Email & Phone Number

Munich, Bavaria, Germany

Developing Next-Level AI Programming Assistants

Moscow, Moscow City, Russia

- built DS team from scratch (5 DS)- created ETL and data quality monitoring (Airflow, Spark/Hadoop)- launched 6 production solutions (real-time & offline) (Python, Go, Lightgbm, Spark, Airflow, Postgres). Used marketplace data about 30+M customers and 80+K sellers- proposed innovative approaches to the loan approval process, based on ML, which significantly increased the company’s revenue and profit- developed pipeline for effective feature selection for models using parallel computing in ray or yarn cluster. Models have less than 50 features each but high accuracy and good stability over time.Projects - Developed ML models for BNPL loans - risk scoring, income prediction, re-scoring * Approval rate increased by 1/3, average limit increased by 3 times but NPL level significantly lowered * Proposed new application stream with approval application without any client’s documents based on ML models only. Conversion to order increased by 50+% while risk remained at base level. * Proposed a pilot for offer fully approved loans for good marketplace customers without loan applications and any documents. Results - acquired addition clients with good risk level.- Developed ML models for loans to sellers * Created a fully automated process, based on ML, for approving loan applications from sellers (SME) without manual underwriting. We don’t ask sellers for any financial statements, etc and make decision immediately * Model used only internal marketplace’s data without credit history, financial statements, etc- Developed solution to assess sellers for factoring product- Developed solution to prevent fraud with cask-back premium points - reduced fraud level by 80+% with very low false positive rateLeft company because of the war

Moscow, Russian Federation

Results - developed 2 online production ML-based solutions with direct profit for the company:- customers segmentation and scoring- antifraud for p2p transactions (payments for crypto and drugs) with good cover and low false positive rateResponsibilities:- defined the scope of ML tasks with business and partners- analyzed data sources (billions transactions). Hadoop, Oracle, Postgres, Informix- created ML models. Python, PySpark, Lightgbm - developed general architecture of production solutions- developed production DB (Postgres) and all calculation scripts for data preparation- development of applications, testing and performance tuning

- Took 30+ Coursera courses- Participate Kaggle and other ML competitions. Kaggle Grandmaster. World ranking top-50.. 06.17 Kaggle/ Sberbank Russian Housing Market - 1st place. Task - to predict realty prices.. 08.18 Kaggle/ Home Credit Default risk - 3rd. To predict repaying bank loans.. 03.19 Kaggle/ Santander Customer Transaction Prediction - 4th. To identify who will make a transaction.. 02.19 Kaggle/ Elo Merchant Category Recommendation - 5th. Help to understand customer loyalty.. 11.17 Kaggle/ Porto Seguro’s Safe Driver Prediction - 7th. To predict auto insurance claims.. 12.16 Kaggle/ Santander Product Recommendation - 7th Bank products recommendation.. 05.18 Kaggle/ TalkingData AdTracking Fraud Detection - 8th. To detect fraud clicks for mobile app ads.. 01.20 Kaggle/ Data Science Bowl - 14th. Uncover the factors to help measure how young children learn.. 09.19 Kaggle/ IEEE-CIS Fraud Detection - 18th. To detect fraud from customer transactions.. 01.18 Kaggle/ Favorita Grocery Sales Forecasting - 21st. To predict sales for a large grocery chain.. 08.17 Kaggle/ Instacart Market Basket Analysis - 25th. Which products will consumers purchase again.. 06.16 Kaggle/ Expedia Hotel Recommendations - 27th Which hotel type will customers book.. 03.17 Hackerearth/ Machine Learning Challenge #1 - 1st. Predict loans default probabilities.. 05.17 AAIA'17 Data Mining Challenge/ Helping AI to Play Hearthstone - 3rd. To predict game outcomes.. 11.19 Zindi Sendy Logistics Challenge 2nd. To predict time of arrival for deliveries in Nairobi.Kaggle profile https://www.kaggle.com/johnpateha

Moscow, Russian Federation

- developed scoring models - implemented modeling with gradient boosting to bank's scoring process

- Project management.- Development of budgeting methodology for automation purposes, design and implementation IT-solutions (based on Oracle Hyperion Planning), quality control. Projects for Gazpromneft, InterRAO UES

Moscow, Russian Federation

- Department’s revenue has increased 5 times, stuff – up to 30 specialists.- Plan and control of the department's projects. Promotion of department's and company’s services.- Manage and participate in key projects.

Moscow, Russian Federation

- Developed methodology of budgeting and costing, implemented budgeting systems- Successfully completed projects.

Krasnodar Territory, Russian Federation

- Responsibility for management reporting, internal control, accounting methodology.- Developed the information system of management accounting and reporting (Delphi, Interbase)

Economics

Machine Learning, Data Analytics

Machine learning, Mining massive datasets (Stanford University), Data Science (9 courses), Statistical Reasoning for Public Health 1&2.