Freelancers › Владислав Бандурин

Developed a chat-bot that could recognize certain text fields in a document from a photo. This gave a huge leap in productivity and
saved a lot of time and money for the company in which I worked.

Technologies used: CNN, OpenCV OCR, Tesseract, Google Cloud, Heroku, Telegram bot API

Developed a large scale recommendation system for a large online store from
scratch. The developed system is capable of solving the all common RecSys business tasks.

Technologies used:
LightFM, XGBoost, Pandas, Flask, Docker, Word2vec, Elastic Search, Dynamodb, AWS S3.

Пользователи могут делать онлайн бронирование и предзаказ через бота с помощью удобного меню, администратор заведения может принимать или отвергать бронь.
Архитектура бота построенна таким образом, что позволяет выдерживать высокие нагрузки. В качестве сервера был выбран AWS, для базы данных - Heroku.

The task was to build a multi-headed a model that’s capable of detecting different types o
sentiments in Russian VKontakte posts. For this task RuSentiment dataset was used. RuSentiment is currently the largest in its class for Russian, it contains 30,521 posts and wich are divided into 5 classes.
To solve this task I use pre-trained vectors for russian language trained on Common Crawl and Wikipedia using fastText.
The best performance was RocAUC score of 0.88 in 5-class classification on the test set. It was achieved by a CNN classifier with in-domain FastText embeddings.

Technologies used: CNN, RNN, Logistic Regretion, fastText, Tf-idf, Word2vec