Retail customer detection and tracking system based on CCTV

Proof of concept of a detection and tracking system done for a worldwide beauty retailer. The system is based on deep neural networks and consists of a customer/staff recognition system, tracking, and action classification subsystems. The resulted system is used for advanced analytics using CCTV in stores. Participation: • Implemented and fine-tuned online tracker for multiple-object tracking problem • Implemented scoring and evaluation pipelines for the system • Augmented pre-trained models for people re-identification process of tracking

COVID-19 epidemiological model and sales impact analysis

A successful take on coronavirus spread modeling and sales impact analysis for a global sports retailer. We had brought theoretical results from the paper to a working state in a very short period of time. The resulting work impacted the decisions of top managers during the initial coronavirus outbreak. Participation: • Created and implemented the COVID-19 response model based on SIR epidemiological model via regression analysis and a system of differential equations. • Implemented country-wise sales impact models and data pipelines for retail and digital stores during the initial lockdown period in Europe

Customer segmentation for beauty retailer

A data discovery project for a beauty retailer. The project was a collaboration work of data science and user experience teams. We have applied machine learning algorithms to big data. The results were a documented data science environment on Microsoft Azure full of data reports with valuable insights. Participation: • Created data loading and preparation pipelines for further analysis using PySpark. • Made a number of analytical lab reports based on transaction data. • Clustered customers using machine learning algorithms into segments using PySpark ML module. • Analyzed correlations between geographical data and customer behavior.

Digital fashion mobile AR app

Mobile app powered by AR and deep learning / digital fitting room. A user can try new clothes or styles using only a smartphone. Participation: • Researched papers on related tasks solved via neural networks • Implemented model and pipeline for human pose estimation task • Realised cross-platform postprocessing library in C++, implemented Python binding for the library