Developing ML microservices to support ride hailing efficiency. This includes case study, implementing ML pipelines (data collection, training and validation), deploying ML models to production, A/B and switchback testing. A list of project that I worked and still work on includes (but not limited by) demand prediction, which helped reducing redundant supply in a number of airports, discount personalisation, which helped optimising discount campaign budgets, and dispatch optimisation, which is aimed at providing the best (the fastest, in most cases) driver for each user.

Working on R&D projects of different types, such as smart baby monitors, optimisationin air traffic control, time series prediction, etc. Preparing proposals for European grants andbusiness accelerators. The company takes part in a couple of international consortia applyingfor European grants.

Research, software development and management in a number of R&D projects, including international SAFE project for non-invasive patient monitoring system and video drone detection system SkyPatriot. Configuration of machine learning pipelines for the projects.Trained a CNN-based pose detection algorithm for IR sensors (used for privacy reasons) and developed software part of the corresponding embedded system. Managed data collection, annotation and validation for SAFE. Participated in hospital trials in Lancaster. Trained an algorithm based on recursive neural networks capable of detecting drones in video stream coming from a 4K camera.Developed algorithms for document classification system (pictures, scans of documents, like IDs, driving licences, etc.). Implemented and optimized video tracking algorithm for an integrated video surveillance system. Implemented and optimized image matching algorithm based on RANSAC keypoint matching. The system helped the customer to filter large information flows to find objects of interest. Implemented OpenTLD algorithm (originally in MATLAB) in C++, optimized it by moving its parts to GPU (CUDA). Enhanced the original algorithm to work under difficult conditions, like multiple look alike objects or low quality video stream.Participated in FP7 project SVETLANA, where developed methods for one-class time series classificationfor flight anomaly detection. Also developed algorithms for ECG analysis. Our work resulted in a number of scientific papers presented at international conferences.