Find the real product decision.
Separate the stated feature request from the user action, business constraint, evidence requirement, and tradeoff the company can accept.
Principal / Staff Product Manager · AI platforms & systems
I make AI ambition .
Fourteen years across Amazon, marketplaces, fintech infrastructure, and shipped AI products. Currently open to Principal/Staff PM and AI Product Lead roles.
14 years product leadership
1,600 NA and EU users
$16B+ incentive portfolio
Enterprise AI Platform
One control plane for how AI is accessed, evaluated, grounded, released, and operated.
Selected work
Three products show how I lead.
Three shipped or deeply validated products show the range. The platform cases underneath them show how I turn repeated constraints into shared capabilities.
Shipped enterprise product · Workflow walkthroughExplore workflow →
Amplify
A reporting request became an AI seller-workflow system embedded where 1,600 commercial users already worked.
89%faster prepAnonymizedworkflow walkthrough−99%inference cost
Live production productRead the case →
SproutRoute
A hybrid travel planner that keeps verified facts, policy, and family-safety data outside generated prose.
350+tests in CIGemini+ Google PlacesMCPagent path
Implemented native productRead the case →
Sprout Math
A native learning product where local state, no accounts, and constrained scope are the trust strategy.
2,400+problem templates2,500+audio clips0accounts required
Local execution / Control planeGoverned LLM Gateway
Arbitrary envelopes exercise DLP, residency, quota, tenant cache, routing, and fallback without fabricating a model answer.
Run locally → Local execution / GroundingRAG GuardrailsQuestions change ranked documents, extractive answers, citations, groundedness, and abstention.
Run locally → Artifact replay / Release integrityAutomated Red-Team GateInspect all 40 committed PyRIT-shaped probe rows, redacted evidence, artifact hash, and BLOCK semantics.
Inspect evidence → Artifact replay / Runtime evidenceAgent ObservabilityFollow a committed forced-failure trace and inspect the 150,000-span collector report.
Inspect evidence → Artifact replay / EvaluationEvaluation Evidence ExplorerModels, dimensions, latency, cost, safety floors, and recommendations come from versioned result artifacts.
Inspect evidence → Local execution + artifact replayAI Safety AuditRecalculate fairness locally, import browser-only CSVs, and inspect a 540-case-per-model provider benchmark.
Run locally → Implemented / GroundingGraphRAG Ontology WorkbenchSource profiling, approved ontology compilation, safe recipe routing, and 80 held-out cases across two domains.
Inspect evidence →How I operate
I start with the decision the company has to make.
Hiring?
Hiring a Principal or Staff PM for AI platforms?
I'm the fit when the role sits at the intersection of AI systems, platform strategy, marketplace economics, and 0→1 execution — and the product has to earn adoption.