Engineered and supported several projects for global brands in Europe and North America • Implemented statistical analysis of the website events to identify web features (floodlights) that drive the business with the ac‑ curacy of 80% using SQL, Catboost Regression and Shapley values. • Extracted insights from media ad banners with Computer Vision techniques by using open source APIs • Successfully managing and analyzing large datasets on BigQuery/GCP of over 100 million records

-Improved the current spelling correction language model in both languages (Kazakh and Russian) to 10% and 12% WER rate (Word Error Rate) respectively using the beam search technique. -Created a text emotion recognizer with an 87% accuracy rate using Rasa. Currently, the model is being used by the company to analyze customers' behavior. -Built a Named Entity Recognition system that classifies 11 entities with a 75% accuracy rate. -Constantly making changes and coming up with improved features. Actively testing and experimenting with SOTA NLP architectures. Currently working on Audio Speech Emotion Classifier.

-Researched and developed NLP models to solve problems around brand safety & implementing various ways of text classification (sentiment, emotion detection) via machine learning, statistics, and optimization. -Built an emotion classifier from scratch using Python with BERT NLP framework resulting in a 46% increase in prediction accuracy. Currently used in Watchout!AI frontend demo app. -Implemented the above ML model as a microservice and integrated it with the demo application using Python. -Developed a sentiment classification model using ru‑GPT with an accuracy of 82. -Provided a data team with a thorough Exploratory Data Analysis on various datasets including Yandex, Sberbank Customer Sup‑ port and used traditional machine learning techniques. -Responsible for preparing datasets for model training and benchmarking of the new and current models.