At Revolut, I own model architecture, evaluation, experimentation, and production ML logic for a customer intelligence platform processing approximately 100K conversations daily. I designed a self-service workflow for evaluating taxonomy updates and an analytics agent that helps product teams explore large conversation datasets.
Previously, I developed production information-extraction systems at CubeRM, including an end-to-end pipeline for public-procurement documents. At SberDevices, I improved NLP components for production text-to-speech systems and contributed to moving inference services from Python to C++.

