Director, Product Management, AI Experience Platform
Own multi-year product vision, initiative P&L, capital allocation, and Go-to-Market execution alongside cross-functional engineering, applied science, and design squads building Royal Caribbean and Celebrity Cruises' shared consumer AI platform and enterprise recommendations across Web and React Native.
Multi-Year Vision, Capital Cases & P&L Ownership: Authored the 2026 follow-on and multi-year 2027 capital expenditure cases that won executive backing to unify isolated brand efforts into a shared platform that builds products; own initiative P&L, unit-economics modeling, executive reviews, and staged Go-to-Market rollout.
End-to-End Guest Journey & Compounding Data Strategy: Anchored roadmap bets in real guest friction from initial cruise discovery through pre-voyage planning and real-time onboard Net Promoter Score (NPS) recovery, treating structured guest preferences and context as a compounding multi-brand asset.
Build-to-Learn Discovery & Portfolio Prioritization: Shifted discovery from static spec handoffs to live interactive prototyping across Product, Design, and Engineering so squads test guest comprehension in days, double down on validated workflows, and cut low-impact scope before committing multi-quarter capacity.
Commercial Trade-Offs & Release Governance: Defined measurable launch gates grounded in real contact-center guest scenarios across answer quality, latency, and cost-per-conversation, aligning Applied Science, Legal, Privacy, and Brand teams around zero-regression release readiness and sunsetting underperforming features.
Autonomous Org Design, Blind-Review Hiring & Coaching: Designed the multi-squad 2027 operating model around end-to-end guest outcomes rather than component silos, instituted blind-review hiring across product roles, and coached product managers through promotion to Senior Product Manager.
Runtime Context Engineering, KV State & Prompt Caching: Architected multi-tier state (Write, Select, Compress, Isolate) across distributed key-value stores, Databricks preference priming, multi-thread compaction, and prefix-stable prompt caching, cutting per-conversation LLM inference cost by 12% to 22% at flat latency while enforcing strict KV TTL invalidation and session isolation.
Eval-Driven Working Memory & Model Migrations: Built golden evaluation datasets from graded transcripts and calibrated 14-dimension LLM-as-judge graders in Braintrust and CI; proved an in-session guest scratchpad lifted multi-turn task completion by 10 percentage points over our no-scratchpad baseline at flat latency, gated two frontier model migrations, and cut an underperforming memory layer.
Context Graphs & Live Trace Observability: Evolved static vector RAG into a structured context graph paired with custom agent-trace inspection extensions and human-in-the-loop review loops to diagnose attention dilution from real payloads.
TypeScript Orchestration, Generative UI & Least-Privilege MCP: Ship production TypeScript alongside engineering across schema-validated Generative UI components and non-overlapping MCP and federated GraphQL tools with session-scoped auth, rate limits, and PII redaction.
Schema-Driven Generative UI System: Co-architected the adaptive UI component language across Web and React Native in Figma and TypeScript, turning structured agent tool calls into interactive comparison tables, choice cards, and batched clarifying questions instead of walls of chat text.
Figma, Motion & Live-Code Prototyping: Prototyped voice-first flows, cross-visit trip continuity, After Effects transition studies, and Premiere Pro executive walkthroughs in days with React Native and coding agents so squads test real guest comprehension before staffing sprints.
Working Memory & Calm Latency UX: Designed an in-session guest scratchpad that lifted multi-turn task completion by 10 percentage points over our no-scratchpad baseline on human-graded transcripts at flat latency, while cutting LLM inference cost per conversation by 12% to 22% via prefix-stable prompt caching.
Trace Review UX & 14-Dimension Quality Bar: Built a live agent-trace inspection extension for human-in-the-loop review and calibrated 14-dimension LLM-as-judge rubrics in Braintrust so conversational UX tone, grounding, and layout accuracy are gated on every release.