Emiliy Feldman Email & Phone Number

Berlin, De

Berlin, Germany

- Setting up data preparation pipelines for model training (ClickHouse, Spark, AWS Glue)- Developing infrastructure and pipelines for model training and validation. Docker-based config-driven solution, production pipeline is driven by AWS StepFunctions.- Data quality and data stability verification. Developed pipelines for regular check of consistency between training and inference features, also for data drift (using Evidently) monitoring.- Model quality checks. Monitoring and alerting pipelines for regular checks of different model metrics.Stack: Python, AWS, PyArrow, CatBoost, VowpalWabbit, Evidently, Airflow, Spark, Clickhouse

Moscow, Moscow City, Russia

ML engineer, developer of recommendation systems, the technical lead of a team.- Built recommendation systems from scratch for 2 services: online books and show ticket sales. Increased metrics compared to baseline: +50% of CTR, +500% of reading books.- Developed ETL process for daily updating of customer recommendations.- Took a part in the creation of an inner framework for rapid development and deployment of recommender models.- Designed a real-time recommender system.Other activities: - Mentoring interns.- Interviewing candidates for junior and middle MLE positions for all company teams.- Participating as a speaker at external meetups on recommender systems development.

Moscow, Russia

Worked as a full-stack ML and data engineer.Developed a recommendation system from nothing: - selected an infrastructure: several microservices interacting via REST-API and Kafka, with ClickHouse as data storage.- set up the collection of information about user actions;- implemented ETL-process to save information in the database;- developed and implemented the near-real-time recommendation model. Also developed and implemented services for the moderation of text, images, and videos. Collected the datasets from the internet myself, then fitted NN-based models using PyTorch.

Moscow, Russia

Development and implementation of universal credit scoring models.- Collected data from various sources like APIs and third-party databases.- Developed data transformation processes, e.g., XML files parsing and converting to tables in the MS SQL database.- Increased old Python code speed up to 20x (from 5 days to 5 hours).- Reorganized the interaction between teams, reducing the time of the retro-test process from 2 weeks to 3 days.- Developed credit scoring models for individuals and legal entities. Increased the quality of old models by 5–20% using additional features.- Implemented all models to production, developed some parts of the API.- Developed a system that allows data scientists to implement ML models in production API in 1 hour.

Moscow, Russia

Development of SaaS machine learning models for marketing automation.- Invented the original model for customer segmentation (RFM analysis). This helps to apply an individual approach to different groups of users. - Developed customer churn model.- Rewrote some difficult SQL code for building analytical reports to Python, which led up to a 60% decrease in calculation time and memory consumption.- Implemented several recommendation models. "Popular" and "popular by categories", as well as more complex methods, such as Alternating Least Squares with the original approach to add explicit item and user features.