Stephen Koo

About

Google Workspace AI engineer who owned major AI assistant infrastructure end-to-end: React/TypeScript UI, a Node.js context-aggregation gateway on Cloud Run, and a Go inference layer on GKE serving Gemini. Built and productionized Gmail RAG + agentic workflows with rigorous evals and guardrails, and has a strong track record of measurable impact (latency, engagement, acceptance-rate lifts) and zero-incident migrations using feature flags/strangler patterns across multiple ML teams.

Key Strengths

• Owned and shipped end-to-end production feature across frontend, backend, data, and infra (React/TS + Node gateway on Cloud Run + Go inference on GKE)
• Designed robust testing strategy (unit + integration + canary rollouts) with error-budget-driven monitoring and automated rollback
• Built multi-region observability with custom SLOs and dashboards for P95 latency/throughput/availability (99.99%)
• Delivered measurable product impact: 40% reduction in integration complexity, 25% P95 latency improvement, 15% engagement lift in first quarter
• Productionized AI-native RAG for Gmail drafting with embeddings/vector search; reduced hallucinations via grounding
• Established LLM evaluation approach using held-out datasets; proved value via A/B test (12% higher draft acceptance, reduced editing time)
• Implemented agentic workflow guardrails (user confirmation for destructive actions, rate limiting, scoped permissions)
• Led zero-incident legacy modernization using strangler pattern, feature flags, dual-writes, and response consistency monitoring
• Effective cross-team execution while embedded across multiple ML teams; delivered reusable infrastructure to unblock parallel roadmaps

Experience

Education

Skills

Languages