I lead and build 0→1 and at-scale products in AI and healthcare.

Twin Health's product is a Digital Twin — a living replica of a member's metabolic health that guides them toward better outcomes. Ask Your Digital Twin (AYT) is the agent-of-agents architecture that makes that Digital Twin something a member can actually talk to.

I set the vision for this Digital Twin experience and secured buy-in from the CTO and CPO, which made it one of the company's strategic Big Bets. I now direct the squad that architected and launched Twin Health's first generation of production AI agents through AYT — spanning Nutrition, Labs, Scheduling, Health Progress, and more — working directly and closely with the CEO, CPO, and CTO on the roadmap, and executing it with a 6-person AI Platform squad.

What shipped, in sequence

• AYT Intercept v2 — rebuilt AYT Intercept with multi-intent recognition and richer context from conversation history, so it can route or respond to incoming member messages more accurately.
• Co-pilot Smart Drafts — an AI-drafted response for incoming messages AYT can't fully resolve on its own, handed off to the right care-team role — coaches, providers, RNs, and others. Established the AYT Smart Drafts ROUGE/similarity benchmark (58% → ~69%) as the team's new quality bar.
• 2-way Voice — enabled members to have a natural back-and-forth conversation by voice across all AYT sub-agents.
• New Domains — learned from usage and built new agents for health insights, labs, weight progression, macros, and more.

Earlier in the platform's life

• Direct AYT v1 — built the original version using RAG retrieval and a food recommendations agent, the foundation the current multi-agent architecture grew from.
• AYT Intercept v1 — shipped the first intercept model handling No action required and AYT responses, deflecting 44% of inbound care-team messages before the v2 rebuild.

Before this work, the sensor experience was unpredictable end to end — spotty setup, no way to catch a failing device before a member complained, and troubleshooting that lived entirely in coaches' heads. I rebuilt it as a connected platform: reliable setup, proactive failure detection, and standardized troubleshooting that's now an AI agent.

Reliability, built proactively

• Redesigned the onboarding flow UX, bringing setup reliability from 43% to 66%.
• Built a rules-based automation that detects failing Dexcom G7/Stelo CGMs and triggers replacement before a member complains — recovering device costs and contributing to $3M+ in COGS savings alongside the Dexy Mail Monitor AI Agent.
• Led a 60-day sensor-performance pilot ("7-Star"), reporting daily adherence with executive leadership; findings shaped the roadmap investments.

Foundational work

• Led the migration and integration of new devices — Dexcom G7 and Stelo CGMs, Garmin, Apple Watch, and Medline BP cuffs.
• Authored the sensor eligibility and lifecycle rules table — signed off by the CEO, CMO, and Finance — so every device decision across the member and clinic apps follows the same logic.
• Defined the Bluetooth sync-reliability architecture in 2019 — still foundational to how the member app handles sensors today.
• Built sensor troubleshooting logic for Fitbit and Garmin — decision trees adopted org-wide — then evolved it into SensAI, an AI agent that now resolves device issues conversationally in real time.
• Owned core app & IoT sensor experiences at scale: 50K members and 100K+ connected sensors.

I joined Twin Health in October 2019 as its first mobile PM, when the "product" was mostly manual coaching over spreadsheets. There was no onboarding flow, no sleep or activity experience, and no support function — just a clinical program that needed a digital front door. I built all three from scratch, grounded in direct field research rather than assumption.

Grounding decisions in the field, not the roadmap doc

• Ran in-person, cross-city user interviews across India before writing a single spec — sitting with members to watch how they actually logged meals, synced sensors, and talked to coaches.
• Engaged 150 members directly in shaping the activity, sleep, and breathing recommendation platform, then led cross-functional treatment experiments with 20+ team members that improved health outcomes by 18%.

What shipped

• Onboarding v1/v2 — the original flow saw roughly 28% of members self-complete their intake questionnaire, forcing coaches to chase the rest by hand. The India-scale redesign brought self-completion to 49% and became the foundation for today's U.S. mobile onboarding.
• Sleep Score v1 → v2 — the first version was built on clinical research; a redesign followed once grievance data showed roughly 30% of members were seeing a blank score and only 7% were syncing from the sleep score page at all. The rebuilt framework is still foundational to the current U.S. app.
• Customer support, 0→1 — stood up Twin Health's support function from nothing, later formalized into the global Zendesk rollout.
• Launched streamlined device-sync features that helped drive the app to a 4.5+ App Store rating.