CEO & Co-founder @ Kaizen — Building AI-powered repair intelligence for the collision repair industry
10+ years in product management, with deep experience building search, recommendation, and generative AI products at scale. Ex-Udemy, Amazon, Tesla.
I build AI systems that turn messy, unstructured domain knowledge into deterministic, auditable outputs. My architecture work is informed by cognitive science — treating hallucination as a design constraint, not just a bug.
- Agent Fluency Score — Open-source benchmark measuring how well developer tools work with AI coding agents
- Navigator — AI-powered development platform that integrates AI across the full product lifecycle: discovery, story writing, backlog management, and TDD-enforced coding agents in isolated cloud sandboxes. Built with Next.js, tRPC, Prisma, Claude Agent SDK, and E2B.
- Multi-agent AI systems for deterministic, verifiable domains — where getting it wrong has real cost
- Constrained agent patterns — Six Thinking Hats, Dialectic reasoning, Pre-Mortem analysis as practical frameworks for reliable AI decision-making
- The judgment layer — tooling and frameworks for the upstream decisions that AI coding tools can't make for you
- Why LLM hallucinations might be features, not bugs — and what cognitive science tells us about building better agents
- Decision debt: the hidden cost of moving fast with AI tools without resolving upstream ambiguity
- How cognitive science-inspired design maps onto multi-agent architecture
- The "what to build" gap — execution speed is commoditized, judgment isn't
| Company | Role | What I Built | Impact |
|---|---|---|---|
| Kaizen | CEO & Co-founder | Multi-agent repair intelligence platform | Filed non-provisional patent (Jan 2025) |
| Udemy | Principal PM, AI | Search, recommendations, gen AI, ML platform | $225M annual revenue org (70% of marketplace) |
| Amazon | Senior PM, Prime | Prime Shipping benefit strategy | Worldwide teams; COVID-19 rapid response |
| Macy's | PM, Recommendations | Personalized recommendation system | $65M incremental revenue in 9 months; 500%+ lift in sales/visitor |
| Tesla | Product Owner | Electronic Parts Catalog (epc.tesla.com) | Built from 0-to-1; 20+ country rollout |
I write about AI agents, product development, and the cognitive science behind building reliable AI systems on LinkedIn.
Languages: Python TypeScript SQL
AI/ML: Anthropic Claude API Claude Agent SDK LangChain LlamaIndex Structured Outputs RAG
Backend: FastAPI NestJS Next.js tRPC gRPC
Data: PostgreSQL Neo4j Qdrant MongoDB Redis pgvector Prisma Drizzle ORM SQLAlchemy
Infrastructure: Docker Google Cloud (Run, Pub/Sub, Tasks, Storage) Vercel GitHub Actions
Frontend: React Tailwind CSS Zustand React Query
Testing: Pytest Jest Playwright Pact Vitest
Monitoring: Sentry PostHog Prometheus Grafana
The bottleneck has shifted from execution to judgment. I build AI systems that respect that.