Nathan Hai Vishnevski

12 AI agents coordinated through an Orca routing layer — a LEARN FIRST knowledge protocol, cross-session memory, and five automated quality gates. Started as my own workflow problem; now it's the factory that ships every other project on this page.

A dashboard that reads raw Excel financials and writes the story behind the numbers. One rule I refused to break: code computes every figure, AI writes text only — nobody presents a hallucinated number to finance people.

Their procedures index lived in one shared Excel file that locked whenever two people opened it. I replaced it with an intranet portal — instant Hebrew search, a read-tracking audit trail with CSV export, manager roles — built to run on infrastructure they already own.

Building with a founding team: a multi-agent product where AI drafts and lawyers decide. Hard legal-safety rules — the AI never pretends to be the lawyer — and Hebrew RTL as a first-class citizen, not an afterthought.

Local-first PWA with an AI layer — task analysis, voice capture, gamified XP — that runs my actual life. You're looking at it right now: tailornate.com is Madko's home. Next milestone: accounts, multi-device sync, web-push reminders.

• Built automation workflows fusing multi-database operational data into decision-ready intelligence
• Real-time act-or-wait calls in time-critical windows, under full operational responsibility
• 200 days deployed — while completing a full CS course load

• Embedded with 18 civilian communities — learned each one's world, bridged it to army systems and resources
• Trained and commanded 50 soldiers
• Weekly briefings reaching 1,200+ soldiers

Currently 2nd year

Accepted a year early — program targets 3rd year+.

Where Polaris was built.