IshipAIproducts
peopleactuallyuse

Led the 0-to-1 launch of a population health analytics module for enterprise hospital networks. The module surfaced AI-generated patient risk scores — but early pilots showed clinicians were seeing alerts and ignoring them.

Interviews with care coordinators revealed the issue was not model accuracy — it was explainability. Clinicians could not trust a score they could not interrogate. The UI had no answer to: why is this patient high risk?

Prioritized an explainability layer over building more alert types. Each risk score now showed contributing factors ranked by weight. Drove cross-functional delivery across engineering, clinical SMEs, and GTM while maintaining HIPAA compliance and enterprise scalability.

Improved care-coordination efficiency by 13%. Accelerated compliant feature releases by 9%. Module became the anchor product in Innovaccer's enterprise GTM motion, deployed across 12+ hospital networks.

Owned the subscriptions and recurring payments roadmap for SMB merchants. On the surface, merchants wanted easier subscription setup. Digging into the data revealed a much bigger problem hiding underneath.

Behavioral analysis showed 72% of failed authorization attempts were never retried. Merchants had no visibility into this — they were losing subscribers silently with no way to diagnose or act on it.

Redesigned retry logic and billing rules before touching onboarding. Counterintuitive call — merchants asked for setup improvements, but the data said the real money was in recovery. Validated via A/B tests using Amplitude and LaunchDarkly.

Increased merchant retention by 11% within two quarters. Reduced time-to-first-transaction by 18% for newly onboarded merchants. Drove 8% increase in merchant expansion revenue through pricing tier experimentation.

Built at OpenClaw Hackathon at AWS Builder Loft SF. Most competitive intel tools are expensive or require manual updates. Web scraping plus LLM synthesis could automate 80% of the work — if you solved the signal vs noise problem.

The real problem was not data collection — it was the diff. Raw scrape data was overwhelming. The value was in what changed since last week, not everything that exists.

Built a before/after diff engine as the core feature. Used Apify for scraping, Redis for caching previous states, Claude API for synthesizing what each change meant in plain English. Shipped a working demo in one day.

Presented live at AWS Builder Loft SF. Real-time change detection across 10 competitor sites. Post-hackathon: iterating with a full dashboard UI, Slack alert integration, and automated weekly digests.