At X5 Tech, I contributed to an e-commerce semantic search model that complements Elasticsearch. I owned the training, validation, and test datasets, implemented hard and soft negative sampling, and co-designed a session-level evaluation that showed a 4 pp Coverage@50 gain over Elasticsearch.
At Moscow Exchange (MOEX) — Finuslugi marketplace, I designed an end-to-end personalization system for customer communications. Its four production models replaced business rules, and A/B tests showed +15% CTR and +10% new purchasing customers while reducing unnecessary SMS spend.
At Andata, I built a daily bid-optimization system that sets Yandex.Direct adjustments for audience segments under budget and target-CPL constraints. I validated it on historical and simulated data before it was deployed as a daily production pipeline.
At MSTU “STANKIN”, I developed lecture materials for a machine learning course, ran seminars, reviewed lab assignments, and held student consultations. I’m growing toward NLP and LLM applications.
