Product Manager · 4 years · AI-native enterprise SaaS
Healthcare GRC, clinical AI, and spend analytics: domains where a wrong answer is a liability, not a bug. 97% faster extraction in production. $500K+ ARR shipped. $1M killed on purpose.
Selected work
Every number below is verified. Estimates are labeled as estimates.
~99.9% extraction accuracy across a 13-analyst, ~25,000-quote/year workflow. Zero added headcount. $5–7M savings identified.
Read case → Quantiphi · 0→1 · FDA SaMD · 2023–24Pivoted the MVP from speed-first to interpretability-first after clinical discovery. NLP accuracy 68% → 88%.
Read case → Quantiphi · Discovery phase · 2023PHI/PII compliance blockers plus negative ROI, found in discovery. Recommended termination before a single sprint.
Read case →Structured UX research and iterative A/B testing on enterprise checkout flows.
A 20-year-old, 366-screen legacy platform rebuilt as a modern AI-first web application. The extraction slice shipped separately to production.
Read case → Enterprise Healthcare GRC · 4-week POC · 2026AI drafts regulatory question sets; a two-person editorial team reviews. 98% passed validation (demo run, unvalidated).
Read case → Enterprise Healthcare GRC · Multi-agent POC · 2026Detects, scores, and drafts grounded impact assessments for regulatory changes. Zero fabricated citations by design.
Read case → Personal build · In daily use · 2026Offline-first, zero-backend productivity workspace for running a four-product portfolio. Built solo, used daily.
Read case → Enterprise Healthcare GRC · Solo POC · 2026Built solo with Claude Code, same pattern as quote automation. Stopped in the same meeting a contract clause ruled out AI-generated content.
Read case →Expertise
Analyst shadowing, SOP audits, clinical stakeholder interviews. Scope from observed workflow, not stated requirements.
WSJF and RICE in production use: 12-sprint roadmaps across 3 engineering teams, defended to leadership.
$1M initiative terminated in discovery on compliance and ROI evidence, with the kill memo as the deliverable.
Shipped extraction pipelines with human-in-the-loop validation, confidence scoring, and field-level provenance.
Grounded-or-silent generation, no-fabrication rules, advisory-vs-gating validation. Trust as a product feature.
Edge-case rule sets and pricing decision trees that took extraction from demo-grade to ~99.9% production accuracy.
Compliance content, vendor credentialing, spend management, where the content is the product.
Clinical AI shipped under medical-device software constraints; PHI/PII risk called out before it burned budget.
Every AI output reviewed by an accountable human. In regulated domains, HITL is the contract, not a feature.