At Yandex, I built an AI agent that helps sales managers analyze client history, processing audio through ASR and turning summaries into RAG. The pipeline scaled to 72K calls a day, and developing the agent with LangGraph reduced response latency by 15%.
I also developed a BERT-based website classifier and its production pipeline, including data labeling and model monitoring. A/B tests showed the project generated over $4 million in additional revenue.
At Sber, I developed an AutoML service to predict competitor behavior in the insurance market, with weekly model refitting and an MLFlow dashboard. The service reduced prediction error from 20% to 8%.
Earlier, I built scoring and forecasting models at Sila Vetra, including a sales model that increased offer-to-purchase conversion by over 10%. I work across ML development and AI agents, and I’m currently working on a research paper about AI agents in trading.
