Health AI
YEAR
2026
OVERVIEW
Led the 0→1 design of an agentic Health AI assistant integrated with One Medical within a HIPAA-compliant system — enabling Amazon customers to get health guidance through conversational AI with safe clinical escalation paths.
The U.S. healthcare experience is fragmented — each provider sees only parts of a patient's health puzzle
Nearly two-thirds of Americans feel overwhelmed by the healthcare system; patients delay care because of costs and confusion
Patients make decisions based on generic internet searches that ignore their personal health circumstances, face weeks-long waits, repetitive paperwork, and impersonal disconnected systems
No existing tool could provide personalized guidance grounded in a patient's complete medical history without requiring manual upload from multiple providers
PROBLEM
PROBLEM
The U.S. healthcare experience is fragmented — each provider sees only parts of a patient's health puzzle
Nearly two-thirds of Americans feel overwhelmed by the healthcare system; patients delay care because of costs and confusion
Patients make decisions based on generic internet searches that ignore their personal health circumstances, face weeks-long waits, repetitive paperwork, and impersonal disconnected systems
No existing tool could provide personalized guidance grounded in a patient's complete medical history without requiring manual upload from multiple providers
PROCESS
Co-developed from day one with One Medical's clinical leadership (Chief Medical Officer Dr. Andrew Diamond directly involved at every stage)
Beta tested with select One Medical members since early 2025; clinical team evaluated performance across thousands of synthetically generated conversations spanning clinical safety, emergency response, and compliance
Evaluation framework required the AI to meet or exceed clinician-level performance on safety-critical decisions before deployment
Launched January 21, 2026 in the One Medical app → expanded to Amazon.com and Amazon app in March 2026
KEY DESIGN DESICION
Multi-agent architecture, not a single chatbot — a core agent communicating with patients, sub-agents handling specific workflows (appointments, Rx renewals), auditor agents reviewing conversations in real time, and sentinel agents standing watch with escalation paths into human providers
Three-layer safety guardrails — Layer 1: Amazon Bedrock Guardrails (input content filtering); Layer 2: LLM-based Auditor detecting 5 healthcare-specific risk categories (mental health crisis, sycophancy, malpractice, malicious users, frustrated users) running in parallel adding zero latency; Layer 3: Safety Assistant that takes over and never hands back when thresholds are breached
Complement, never replace providers — clinical escalation automatically triggers when symptoms warrant human judgment, connecting patients via messaging, immediate video call, or same/next-day in-person appointment
Personalized via Health Information Exchange — accesses complete medical records (labs, medications, diagnoses) without requiring patients to manually upload from multiple providers
HIPAA-compliant by default — conversations are NOT automatically added to the medical record; encryption + strict access controls; Amazon does not sell PHI
OUTCOME
~9.4K daily active users — scaled from 0→1 within months of launch
~2K new One Medical memberships/day — users loved it enough to convert to paid members
"Overwhelmingly positive" reception — from both patients and providers, per Amazon's official announcement
Expanded from beta → all U.S. customers — launched on One Medical app (Jan 2026), then Amazon.com and Amazon app (Mar 2026)
Partnerships with major health systems — Rush University Health and Cleveland Clinic integrated for primary-to-specialty care continuity
Health AI
YEAR
2026
OVERVIEW
Led the 0→1 design of an agentic Health AI assistant integrated with One Medical within a HIPAA-compliant system — enabling Amazon customers to get health guidance through conversational AI with safe clinical escalation paths.
The U.S. healthcare experience is fragmented — each provider sees only parts of a patient's health puzzle
Nearly two-thirds of Americans feel overwhelmed by the healthcare system; patients delay care because of costs and confusion
Patients make decisions based on generic internet searches that ignore their personal health circumstances, face weeks-long waits, repetitive paperwork, and impersonal disconnected systems
No existing tool could provide personalized guidance grounded in a patient's complete medical history without requiring manual upload from multiple providers
PROBLEM
PROBLEM
The U.S. healthcare experience is fragmented — each provider sees only parts of a patient's health puzzle
Nearly two-thirds of Americans feel overwhelmed by the healthcare system; patients delay care because of costs and confusion
Patients make decisions based on generic internet searches that ignore their personal health circumstances, face weeks-long waits, repetitive paperwork, and impersonal disconnected systems
No existing tool could provide personalized guidance grounded in a patient's complete medical history without requiring manual upload from multiple providers
PROCESS
Co-developed from day one with One Medical's clinical leadership (Chief Medical Officer Dr. Andrew Diamond directly involved at every stage)
Beta tested with select One Medical members since early 2025; clinical team evaluated performance across thousands of synthetically generated conversations spanning clinical safety, emergency response, and compliance
Evaluation framework required the AI to meet or exceed clinician-level performance on safety-critical decisions before deployment
Launched January 21, 2026 in the One Medical app → expanded to Amazon.com and Amazon app in March 2026
KEY DESIGN DESICION
Multi-agent architecture, not a single chatbot — a core agent communicating with patients, sub-agents handling specific workflows (appointments, Rx renewals), auditor agents reviewing conversations in real time, and sentinel agents standing watch with escalation paths into human providers
Three-layer safety guardrails — Layer 1: Amazon Bedrock Guardrails (input content filtering); Layer 2: LLM-based Auditor detecting 5 healthcare-specific risk categories (mental health crisis, sycophancy, malpractice, malicious users, frustrated users) running in parallel adding zero latency; Layer 3: Safety Assistant that takes over and never hands back when thresholds are breached
Complement, never replace providers — clinical escalation automatically triggers when symptoms warrant human judgment, connecting patients via messaging, immediate video call, or same/next-day in-person appointment
Personalized via Health Information Exchange — accesses complete medical records (labs, medications, diagnoses) without requiring patients to manually upload from multiple providers
HIPAA-compliant by default — conversations are NOT automatically added to the medical record; encryption + strict access controls; Amazon does not sell PHI
OUTCOME
~9.4K daily active users — scaled from 0→1 within months of launch
~2K new One Medical memberships/day — users loved it enough to convert to paid members
"Overwhelmingly positive" reception — from both patients and providers, per Amazon's official announcement
Expanded from beta → all U.S. customers — launched on One Medical app (Jan 2026), then Amazon.com and Amazon app (Mar 2026)
Partnerships with major health systems — Rush University Health and Cleveland Clinic integrated for primary-to-specialty care continuity
Health AI
YEAR
2026
OVERVIEW
Led the 0→1 design of an agentic Health AI assistant integrated with One Medical within a HIPAA-compliant system — enabling Amazon customers to get health guidance through conversational AI with safe clinical escalation paths.
The U.S. healthcare experience is fragmented — each provider sees only parts of a patient's health puzzle
Nearly two-thirds of Americans feel overwhelmed by the healthcare system; patients delay care because of costs and confusion
Patients make decisions based on generic internet searches that ignore their personal health circumstances, face weeks-long waits, repetitive paperwork, and impersonal disconnected systems
No existing tool could provide personalized guidance grounded in a patient's complete medical history without requiring manual upload from multiple providers
PROBLEM
PROBLEM
The U.S. healthcare experience is fragmented — each provider sees only parts of a patient's health puzzle
Nearly two-thirds of Americans feel overwhelmed by the healthcare system; patients delay care because of costs and confusion
Patients make decisions based on generic internet searches that ignore their personal health circumstances, face weeks-long waits, repetitive paperwork, and impersonal disconnected systems
No existing tool could provide personalized guidance grounded in a patient's complete medical history without requiring manual upload from multiple providers
PROCESS
Co-developed from day one with One Medical's clinical leadership (Chief Medical Officer Dr. Andrew Diamond directly involved at every stage)
Beta tested with select One Medical members since early 2025; clinical team evaluated performance across thousands of synthetically generated conversations spanning clinical safety, emergency response, and compliance
Evaluation framework required the AI to meet or exceed clinician-level performance on safety-critical decisions before deployment
Launched January 21, 2026 in the One Medical app → expanded to Amazon.com and Amazon app in March 2026
KEY DESIGN DESICION
Multi-agent architecture, not a single chatbot — a core agent communicating with patients, sub-agents handling specific workflows (appointments, Rx renewals), auditor agents reviewing conversations in real time, and sentinel agents standing watch with escalation paths into human providers
Three-layer safety guardrails — Layer 1: Amazon Bedrock Guardrails (input content filtering); Layer 2: LLM-based Auditor detecting 5 healthcare-specific risk categories (mental health crisis, sycophancy, malpractice, malicious users, frustrated users) running in parallel adding zero latency; Layer 3: Safety Assistant that takes over and never hands back when thresholds are breached
Complement, never replace providers — clinical escalation automatically triggers when symptoms warrant human judgment, connecting patients via messaging, immediate video call, or same/next-day in-person appointment
Personalized via Health Information Exchange — accesses complete medical records (labs, medications, diagnoses) without requiring patients to manually upload from multiple providers
HIPAA-compliant by default — conversations are NOT automatically added to the medical record; encryption + strict access controls; Amazon does not sell PHI
OUTCOME
~9.4K daily active users — scaled from 0→1 within months of launch
~2K new One Medical memberships/day — users loved it enough to convert to paid members
"Overwhelmingly positive" reception — from both patients and providers, per Amazon's official announcement
Expanded from beta → all U.S. customers — launched on One Medical app (Jan 2026), then Amazon.com and Amazon app (Mar 2026)
Partnerships with major health systems — Rush University Health and Cleveland Clinic integrated for primary-to-specialty care continuity