DevTeam.Space Developer Profile, Kirill

For a podcast startup, Kirill developed a system to segment text into semantic paragraphs. Utilizing transformer architectures and various Bert models, he enabled the automatic division of podcast texts into meaningful sections and the generation of titles for these sections.

In this project aimed at detecting starting signals or shots in athletic events, Kirill designed a system architecture employing convolutional and recurrent neural networks. His work focused on enabling the system to distinguish and ignore irrelevant sounds coming from different directions.

Kirill tackled the challenge of identifying traffic violations by creating a system for vehicle detection, multi-angle identification, violation detection, and license plate recognition. This involved exploring various OCR methods to accurately recognize license plates.