Projects
Computer vision
Detector, segmentator and set of rules for determining damage
A model for searching for structural damage on an extemely small volume of images
Deep learning detector, segmentator and classificator for determining bisycles parts
Bisycles parts detection, segmentation and classification
Images annotation for CV project
Bounding boxes annotation, segmentation annotation, and classification annotation
- used open data with segmentation annotation from drone
- build segmentation model to find different classes: people, bicycles, roofs (22 classes)
Segmentation of objects for drone photos
Content video and photo AI system
- deep learning classifier for videos
- face search service with elastic search
- tagging for images ilmplementation of web service for site
Scoring and analytics
Insurance scoring system
- developed a customer deсision system with machine learning models
- transferred modeling to modern open source libraries from commercial software
- models were implemented into production
- we have solved several optimization task for retail industry (find minimum cart with constraints for suppliers and buyers)
- it allows to reduсe cost of buying for small companies
Optimization for retail industry
Graph analytics for insurance company
- graph in python networks
- fraudulent customers were found in graph
- results transferred to neo4j
- neo4j was developed to production mode
- graphSage was part of the progect
- build models deaths prediction (XGBoost, LGBM, optuna, XGBSE, pycox)
- helps to make life longer
Optimization for retail industry
Time series prediction for commodities
- build models for metal prices (pytorch forecasting, prophet, XGBoost, LSTM, greykite)
- make buy and sell recommendations for management
- extracting data from different sources (articles, prices, content)
- automation of scraping and parsing process (airflow, cron)
- ETL process development and data strage
Web scraping
Time series prediction for commodities
- build models for metal prices (pytorch forecasting, prophet, XGBoost, LSTM, greykite)
- make buy and sell recommendations for management
NLP
Article generator for SEO optimization
- build algorithm to generate articles from different domains
- build web service with API
- NLP models
- different transformer models (GPT, GPT2, CTRL, BERT, XL, models)
- we have built url artticles parser
- did different tasks for articles text (topic modeling, named entity recording, sentiment analysis, spam classifier, mistakes, summarization)
- build meta model for text classification
NLP web articles classification problems
Different NLP tasks for Oil and Gas company
- build big inner email communication graph
- did different tasks for email texts (topic modeling, named entity recognition, sentiment analysis)
- extract non formal groups in company based on topics and entities
Our servers available for free on projects
IBM x3960
- 1TB RAM
- 120 CPU cores
- 3 GPU 1080ti
Social projects
Education Course Mail.ru - Geekbrains
- developed and launched the educational course “Machine Learning in Business”
- trained over 150 studens
More than 50 researh videos (data siens, frontend, backend, product management and design)
Data science and development youtube channel
