Alex Korshak

About

Fifteen years building and scaling systems through the layers — from individual contributor to VP of Engineering. Most of that time spent in regulated or operationally serious environments where shipping fast and shipping safely aren’t in tension if you design for both from the start.

These days I lead AI delivery end-to-end: roadmap, architecture, the first build, then the team and operating model that scales it. I care about evaluation discipline, rollback paths, and what happens at 3 a.m. when something drifts.

Selected Work

• AI-fueled GitOps delivery

  Regulated FinTech

  Custom delivery platform with self-healing rollback, flaky-test bucketing, and dynamic config injection. Developer self-service without abandoning guardrails.

• SOC 2 change governance via CI

  Compliance Automation

  Pipeline-derived audit evidence with risk-tier approvals. Low-risk changes auto-approved; CAB bottleneck removed. SOC 2 Type II delivered in roughly nine months.

• ML decision engine

  Applied ML

  Risk scoring that compressed application processing time and auto-approved low-risk cases. Built on internal history plus external bureau signals.

• Edge-first performance rebuild

  Commerce

  Cut page loads from 6–7s to ~3s against unavoidable supplier-API latency. Async aggregation, progressive rendering, near-zero unplanned outages.

Code

• repoarchitecture-notes — short essays on production AI operations
• repoai-delivery-guards — mechanical CI checks for LLM-touched changes
• repoagentic-patterns — bounded agentic workflows with eval gates

Notes

• 001What ‘production-grade AI’ actually means
• 002Judgment-as-code: governance you can deploy
• 003Bounded autonomy — designing agents that fail well

Elsewhere