Waste recycling startup.Autonomous waste bag sorter for city yards.Something beautiful, something special, something memorableComputer vision, roboticsSite: http://en.ce-ecobot.comDemo: https://youtube.com/shorts/JKz8s7dFvHg?feature=share

Uplift modeling, A/B testing, tabular analytics

Stack: Hadoop, PySpark, SQL, sklearn, airflow

Startup.Computer Vision- XRAY-CV. Implemented an algorithm for solving the segmentation problem prohibited items in xray scanner suitcases at airports. The model for the customer surpassed the analogue of competitors,thanks to which the company was able to conclude a contract for a long-term cooperation with several airports.- People Detector. Completed a project on real-time detection of people in 3 positions (standing, sitting, falling), integrated the model on the Jetson Xavier.- Teaser-Analytics. Analysis of the influence of teasers on viewing the premiere on based on TV channel data on second viewing and demographics watching.- Shop CV. Completed a project to highlight clothing and accessories for pictures. MVP allowed the customer to close one of their problems.- Car-Insurance-CV. Created an algorithm to filter bad images quality, which improved recognition accuracy by 2 times in comparison with past decision, for a project to assess damage to vehicles for insurance companies, improved the damage detection algorithm by car. At this stage, the project is at the stage of sale insurance companies.- Mobile-Damage-CV. Completed a damage analysis project various phone models based on their photos.Stack: Python, PyTorch, Mxnet, GluonCV, OpenCV, Pandas, Lightgbm, Sklearn, Docker, CI-CD, MongoDB, TensorRT, Triton.

The largest private bank in Russia.NLP- Created analytical services for the internal needs of the company- Created a call routing solution based on text classification and NER

The largest telecom operator in CISNLP, Marketing analytics- Automatic clustering of samples from social networks by topics. Developed an application that can split the source text from social networks into topics, highlight the pros and cons of products, get statistics onnamed entities, etc. The application is regularly usedmarketing analysts of the department for ad-hocresearch. The time of work of one person was reduced from 5-6 hours to 0.5-1 hour for this type of task.- Marketing and customer analytics of usersIdentification of relevant and toxic services and key marketingcompany metrics.Stack: Python (Pandas, Numpy, Sklearn) + NLP stack (Spacy, NLTK, PyMorphy),SQL, Docker

The largest marketplace in CIS.NLP- Sentiment analysis of user commentsThe result of the work is a web service (API) that allows predict the product rating based on the user's text comment. Tested and implemented various ML algorithms for NLP: word2vec, doc2vec, LSTM, BERT (SOTA), Knowledge distillation. Cleaned and analyzed the data, chose methods and approaches to solve the problem. API actively used in the company.- Decoding fashion context using word embeddings. Research project to create a universal vector productionusers based on their purchases and site activity versed in scientific articles on the subject, implemented algorithms.Stack: Python (Pandas, Numpy, Sklearn, Ligbtgbm, Catboost, PyTorch), SQL(Clickhouse), Docker, Kubernetes, Linux.