Every product is a sequence of decisions members make in seconds. I design the seconds.
Behavioral psychology plus AI systems, at 40M+ member scale.
Three decisions, designed.
Personalization at 40M+ scale
I own the surfaces where ML ranking output becomes member-facing product, and the behavioral signals that flow back into the models.
Fewer doors, faster walking: the +20% lived in what we removed. New members picked one intent instead of browsing everything — Hick's Law, applied at the first screen.
Ten interventions, one dealt forward per member per moment. The 20% is in the choosing. Keeping the reward unpredictable is what held the lift — a variable reward schedule.
GenAI Promotion Engine
Strategy first: when to reinforce a member versus win one back. Then the full spec, event schema, prompt contracts, and API input/output, across 10 intervention templates.
Built and directed the growth team behind a multi-channel engagement system: $20M annual revenue, 15% churn reduction.
Built with AI, end to end
Four-agent pipeline (Orchestrator, Copywriter, Nudge Assistant, Designer) with explicit prompt contracts and handoff logic between agents.
Consumer meal-planning app on a two-model Claude architecture with Supabase; designed, built, and deployed solo.
chefbuddy.appExplicit contracts between agents, so each handoff is a decision, not a hope. Each agent gets a defined input, output, and handoff rule before it runs — that is what makes a four-agent pipeline debuggable instead of a black box.
Email Analyst to Senior Product Manager.
Four promotions in under 7 years at DISQO, then the move to Prodege.
Each step traded a bigger surface for a bigger question. The throughline: hired at one level, performed at the next.
Each promotion came from owning the next problem before it was assigned.
How the seconds get designed.
Production A/B with guardrail metrics on every experiment. Event schemas and API contracts I write myself, then work through with engineering.
Hook Model, endowed progress, variable rewards. Behavioral craft that makes surfaces feel human, not mechanical.
What I'm building toward: model-driven consumer surfaces. Recommendation, personalization, and agentic experiences built to respect how people actually decide.