Open to product opportunities
Product Manager · Data, AI & enterprise workflows
I build data products for regulatory researchers, enterprise teams and carrier operators.
Five years from discovery to rollout.
$40M+ in investment decisions informed. $1.5M enterprise contract secured with the team.
Tolendi TastybaySan Francisco Bay AreaProduct choices in regulatory research, enterprise software and carrier operations.
I used rules for clear records and routed ambiguity to model-assisted human review.
Strength conflict → human review
I prioritized workflow fixes, then narrowed executive scope to reporting on shared data.
I replaced dashboard-first direction with reconciliation and evidence-linked alerts.
Invoice open. Payment referenced.Inspect bank and factoring evidence.
Awaiting operator review
From a regional pilot to nationwide coverage across 6,000+ villages.
Discovery, delivery and adoption across consequential workflows.
Jun 2026 — Present
Product Management
Turning fragmented carrier records into evidence-linked operational alerts.
Related caseSep 2025 — Jun 2026
Product Manager · Graduate Practicum
A deterministic-first FDA reconciliation workflow with human exception review.
Related caseJun 2023 — Aug 2025
Product Manager II
Enterprise finance, approvals and reporting, from business case to regional rollout.
Related caseJun 2021 — Jun 2023
Product Manager
Explainable infrastructure priorities, from a regional pilot to national coverage.
Related caseI grew up on a farm in a village in Kazakhstan. Later, fieldwork showed me how internet access, digital literacy and local capacity changed the same workflow. At CSI, that meant piloting the submission process before scaling it.
More about my approachInspectable modeling projects, with offline results and honest validation boundaries.
A prototype for financial reviewers to screen distress risk, inspect drivers and explore revenue/cost shocks.
Logistic regression retained for explainability over boosted challengers. The public repository reports 540,860 filing-year rows, a 145,852-row core Form 990 modeling panel, and 0.810 ROC-AUC on the 2022 holdout. These are proxy labels and offline evaluation, with no deployed user-impact claim.
A working similarity model for scouting analysts to compare playing styles and explore replacement candidates.
The linked repository includes position-based normalization, PCA and cosine KNN, with scree and silhouette/elbow plots for structural validation. Market-value and future-upside models remain roadmap items in this version. The similarity model has not been validated through a club deployment or scouting outcome.