Sergey Zotov Email and Phone Number

* I developed a system for training neural network classifiers of types and various categorical parameters of goods by their SKU (text description) and other categorical and numerical features, including a text preprocessing pipeline. During my work on this project, I achieved the following results:1. An accuracy of trained classifiers of product types by SKU reached 96%2. The system used a large number of data rows to train classifiers - up to 2.2 million lines - and did not overload the PC's RAM3. The system was built in such a way that any user could work with it without having any special knowledge — so that it was enough to set a few settings, specify the location of the data and start the learning process, and then get a classifier and use it in practice4. Classifiers work is distributed among processor cores to speed up preprocessing of data for training classifiers or classification of rows trained by classifiers. This significantly reduces the time of training and working of classifiers with real data* I was involved in developing, supporting and optimizing a company's internal API, which provides access to a series of tools for quickly working with data and parsing data from the Internet* I developed, supported and optimized microservices in Go for the company's internal API* I increased the speed of data processing tools via the internal API by several times, and sometimes by orders of magnitude, by translating them into the faster Go language and distributing the work among multiple processor cores

* I applied classical machine learning methods to classify pixels of multispectral images using the Scikit-learn package* I compared the results obtained by different classification methods* I proposed and tested methods for preprocessing multispectral images to increase classification accuracy