Context

A significant volume of content with racist undertones was circulating in Russian-language media, encoded through slang and euphemisms specifically designed to evade standard keyword-based moderation.

Challenge

Keyword filters couldn't keep up: the language kept shifting, and the coded terms were nearly impossible to catch without understanding the underlying context and subcultural codes.

Result

Osavul's AI models identified and catalogued a vocabulary of coded racist terms used across Russian media, exposing the real scale of the phenomenon — a finding later echoed in independent reporting on hidden racial slurs surfacing across Russian media.

Discuss a similar challenge