Case Study  ·  George L. Reynolds  · georgelreynolds.com

Ask FlyteHealth

EMR-integrated behavioral health AI — prototype to production at FlyteHealth

Role
Sr. Director of Product, Patient Experience
Organization
FlyteHealth
Status
Production — Deployed
EMR
Direct athenaOne chart access — medications, vitals, instructions
RAG
Strict gating — no open-web fallback, no speculation
HIPAA
HIPAA & HITRUST compliant infrastructure on AWS

A clinical environment where errors have real consequences

Ask FlyteHealth is a HIPAA- and HITRUST-compliant, EMR-integrated AI assistant supporting patients across onboarding, nutrition coaching, exercise guidance, medication adherence, blood pressure monitoring, and personalized care plan generation.

Operating with direct access to medications, vitals, provider instructions, and longitudinal patient context.

Product Screens — Production

EMR-grounded responses in a live clinical environment

Each response is anchored to real patient data — appointment summaries, biometric trends, and curated clinical content — not generic health information.

screenshot

Progress response grounded in biometric data — blood pressure, BMI, activity logs

screenshot

Behavioral coaching response from curated content library

screenshot

Post-appointment summary pulled directly from the EMR visit record

Purpose-built for clinical context retrieval

The system pulls context from three sources — the EMR, the data lake, and a curated content library — and routes it through a retrieval-grounded generation pipeline with deterministic risk controls.

Orchestration & Models

  • Python + LangChain orchestration
  • Anthropic models via AWS Bedrock
  • FAISS-based knowledge base retrieval

Clinical Data Integration

  • Dynamic patient data via FlyteHealth data lake
  • EMR chart access — medications, vitals, instructions
  • Curated content library

Frontend

  • React (web)
  • React Native (mobile)

Infrastructure

  • HIPAA & HITRUST compliant
  • Secure patient data handling
  • Audit-ready architecture

Evaluation embedded in the architecture

Evaluation is not a post-deployment audit process — it is built into the development loop to allow rapid iteration without increasing clinical risk.

Control Purpose
Strict RAG gating No open-web fallback. All responses grounded in EMR, data lake, or content library.
Medication framing constraints Responses involving medications are constrained to factual, non-prescriptive framing.
Structured output validation Outputs are validated against expected structure before delivery to the patient.
Deterministic rule enforcement Certain response categories are governed by hard rules, not LLM judgment.
Automated regression testing Prompt version changes are tested against a regression suite before deployment.
SME-guided clinical review Subject matter experts review outputs for clinical appropriateness and tone.

George L. Reynolds

Health Technology / Digital Health Product Leader with 20+ years of experience building and commercializing clinical-grade digital health products. Ask FlyteHealth was led from prototype to production release as Senior Director of Product, Patient Experience at FlyteHealth.

georgelreynolds.com  ·  LinkedIn