ambiguous example dosage form and application conflict
rules inconclusiveLLM fallbackhuman review
02data product · recommendation system · public sector
infrastructure prioritization and recommendation system
transformed fragmented regional data and more than 60 infrastructure standards into a recommendation system that helped government teams prioritize investments across thousands of communities.
4,000+ communities60+ standards in product logic$40m+ allocation decisionsmultiple categories and regions
all regionswater + schools + roadsbudget constrained
ranked projectregionscore
01water pipelineregion a92
02rural schoolregion b87
03road rehabilitationregion c81
why #1high service risk
critical standard gap across 14 communities with feasible delivery inside the current budget.
$8.5m allocated · $4m remaining
03data product · analytics · product optimization
regional payroll reporting and analytics platform
improved the accuracy, performance, and adoption of a regional payroll reporting platform used across schools, universities, mayoral offices, and other government organizations.
before · slow report retrievalafter · 50% faster queries
02
currently building
work in progress, with assumptions and evidence kept clearly separate.
ai product · logistics · 0-to-1MVP in progress
ai operations platform for small trucking carriers
designing an AI operations product that turns invoices, emails, bank transactions, factoring documents, and operational records into financial alerts and recommended actions for small trucking carriers.
validated problems
documents and transactions are fragmented across email, spreadsheets, paper, and financial systems.
current assumptions
explainable payment and concentration alerts may create more value than a general assistant.
MVP functionality
invoice matching, revenue alerts, broker concentration, workflow audits, and recommendations.
planned pilot metrics
matching accuracy, resolution time, alert usefulness, and repeated use.
next experiment
pilot invoice-payment matching with small carriers and observe whether alerts change action.
invoice is past the expected payment window and no matching bank transaction was found.
invoice parsedbank checkedrisk flagged
recommended actionverify factoring status and follow up with the broker
03
beyond work
projects i built to explore problems, datasets, and subjects i genuinely care about.
prototype
nonprofit financial shock simulator
a Python and Streamlit product using more than 1.7 million nonprofit records to identify organizations vulnerable to financial shocks.
The product transformed a large public dataset into an interactive decision-support tool for exploring financial vulnerability among nonprofit organizations.
1.7m+ nonprofit records75% precision in highest-risk segments
build a six-month liquidity plan and diversify grant revenue.
working model
football scouting and recommendation system
a data-driven player similarity and scouting system built from ten years of football performance and market-value data.
The product identifies similar players, potential replacements, undervalued players, and future-growth candidates.
250k+ recordsten years of player data
player recommendations
8
selected midfielderage 24 · market value $4.2m
progression
pressing
chance creation
recommended replacements
player a91% similar$2.8m · undervalued
player b86% similar$3.4m · growth candidate
personal photograph
04 · about
a path shaped by curiosity
i’m originally from Kazakhstan and currently based in the San Francisco Bay Area. my path into product was not something i planned from the beginning. i started in financial audit, where i quickly realized that i was more interested in improving the work than repeating it. i kept looking for ways to automate routine processes, simplify how financial data was handled, and remove work that did not need to be manual.
that curiosity gradually pulled me closer to product management. i wanted to understand the full problem, work directly with engineers and users, and see whether something i helped build actually made a difference. product gave me the opportunity to move from improving isolated tasks to building systems used by real people.
while working on decision-support tools and a recommendation system, i became increasingly interested in what was happening behind the product: how the data was structured, how recommendations were generated, and how machine learning could improve decisions. i wanted to understand that side of the work more deeply, which led me to pursue an MS in business analytics at UC Davis.
today, i’m continuing that path into AI products. i’m especially interested in building tools that combine data, automation, and human judgment to make complicated workflows easier and decisions clearer.
outside work, football takes up an unreasonable amount of my attention—the version mostly played with feet. i also enjoy exploring new places, learning how different systems and cultures work, and occasionally building football scouting models that confirm professional scouts probably know more than i do.