Experience
11+ years in product, told in numbers
From fintech to marketplaces: AI assistants, ML products, edtech. Every role through a measurable outcome — not a list of duties.
0.27% → 26%penetration in one year: share of active sellers in training (140k MAU) — at zero acquisition cost
₽15Bsized the TAM, showed leadership an untapped distribution channel — WB’s own seller base — and secured the PRO Wildberries budget
−60% FTEper course with higher quality: moved content production to LLMs — generation, RAG, agents, auto-graded homework; completion above 60% vs a 35–45% market norm
ROI 120%in the first paid cohorts of live training; shut down a RAG-based AI-assistant pilot within a month when seller metrics didn’t move, and redirected resources to live courses
DiDdifference-in-differences on trained vs untrained seller cohorts showed training lifts retention and cross-sell of paid products — made that link the product’s target metric
+40%organic traffic to the training site in 3 months: launched the WB Insight brand media in 2 weeks
12% → 50%bot automation rate ×4 in 8 months: share of requests resolved with no agent, at ~10k questions a week
68MSber customers reached by the mortgage skill in the Salute assistant; took it into the top 10 in 4 months, first among the tribes
+24%conversion to owner contact (4.2% → 5.2%): ML property picks on the domclick.ru home page — collaborative filtering, offline scoring of hundreds of thousands of listings, A/B against the heuristic
RoBERTamoved the intent classifier from fastText to RoBERTa; introduced a resolved-request criterion, labeling, and per-intent thresholds
0.02% → 0.9%referral-program penetration ×45 in 3 months after the relaunch, on a 700k-customer base: two-sided bonus, invite by link, points with no bank details or tax paperwork
Launch“Alfa-Security” subscription from zero in a year — on sale since 2021: employee background checks for hiring, copyright-risk protection
P&Ldefended the roadmap and product economics at the finance and product committees
6 mopayback of the AI support bot on a 60k-customer base: end-to-end scenarios where the bot acted in the account itself — opening an account, for instance
−43%median resolution time (30 → 17 min) in 3 months: end-to-end scenarios and direct routing to tier 2 at 98% routing accuracy
×20faster bot-scenario launch (1–2 weeks → hours): a hub where support staff wrote Lua scenarios themselves, no developers needed
fastTexta model hierarchy for long-tail questions and an agent copilot for below-threshold intents
Klass365 (Aktion Group)
Project Manager
Jun 2014 — Jun 2017
SaaS for retail, e-com, and services: management and inventory accounting, POS, and CRM. Owned service quality and the release plan, led the QA team, rolled the product out to clients. Took release frequency from one every 3 weeks to two a week — regression on Selenium IDE with a team of 3 testers. Ran 40+ client rollouts in 3 years and launched a 70+-article knowledge base and a helpdesk.
Education
Kharkiv National University of Radio Electronics, 2013 — applied mathematics.
Skills
Product managementProduct strategy & roadmapLLM assistants & chatbotsRAG, agents, prompt engineeringScoping work for DS teamsA/B testingUnit economics & P&LCustDevCJMUX/UI designBacklog prioritization (RICE / MoSCoW)
Tools
PythonSQLGrafanaPower BIClickHouseAmplitudeFigmaMiroJira
Languages: Russian — native; English — B2.