Project: motion prediction
What was done:
As part of the work, developed and modified neural network models to predict the behavior of road users.
Conducted experiments to improve the efficiency and accuracy of the model.
Introduced a number of new approaches from research articles in the field of motion prediction.
Created new analytical tools including metrics development and data analysis system.
Organized and tuned the data cleaning, preparation and collection pipelines, automating the collection processes and increasing the dataset size by two orders of magnitude (Spark, Dagster).
Developed a methodology to evaluate the utility of each sample in the dataset, which simplified collection and improved the quality of training on new data

Project: for Beeline
What was done
Developed an algorithm to search and diagnose problems at base stations of a telecom network covering the entire country. The algorithm identifies various anomalies and degradations such as internet traffic problems, increased load and emergencies using over 30 different metrics to analyze.
My main contribution to the project was the complete development of the model's pipelines: from offloading extensive data from various sources such as Oracle, MSSQL, Hadoop and PySpark to creating the analytical model. I led the project from the development phase to a successful deployment to a working environment using modern tools and technologies such as Gitlab CI/CD, Docker and Airflow.
Results: Reduced the number of requests (trouble tickets) by more than 2.5x. Increased troubleshooting response rate by 30%