Duke Morgan  /  Clinical AI Architect • AI Enablement • Director of Clinical Systems

Where clinical reality
meets production AI.

I'm a licensed clinician (RN) and Director of Clinical Systems who architects governed AI platforms, multi-agent RAG workflows, and enterprise AI enablement programs. From 600+ table clinical database intelligence and BAA-governed Azure/OpenAI/Claude agent fleets to reusable operational automation, I build resilient systems grounded in real clinical workflows.

Bridging bedside clinical intuition, enterprise AI governance, and hands-on systems engineering to turn complex operational problems into governed, repeatable capabilities.

Explore Architecture & Case Studies View Résumé (PDF) LinkedIn
Duke Morgan, Director of Clinical Systems
RN  /  Multistate Compact  /  Active
600+ Clinical SQL Server tables in governed schema RAG
Dual-Lane BAA Azure AI Foundry & zero-PHI model routing
3M+ Clinical records unified across 4 legacy EMRs
Multi-Agent Cross-model verification & reusable skill library
20+ Yrs Emergency RN practice to Clinical Systems Director
01 / Clinical AI & GovernanceMake complex clinical systems safe & navigable

600+ table EMR schema intelligence, BAA data boundaries, and PHI-isolated model routing.

 02 / Reference ArchitectureInteractive Multi-Agent RAG blueprint

Explore 4-plane ingestion, hybrid dense/BM25 retrieval, agent state graphs, and faithfulness guardrails.

 03 / AI Enablement & AgentsScale reusable skills across teams

Conversational workflow intake, persistent session memory, and cross-model review gates.

 04 / Credentials & Track RecordClinical licensure & frontier AI

Active RN Compact license, BSHA, plus OpenAI, Anthropic, Microsoft Azure, and Google credentials.

01 / Clinical AI & enterprise governance

AI that helps solve the work around care.

My clinical AI work starts with the systems clinicians and analysts rely on every day: navigating complex data models, investigating workflow bottlenecks, establishing the enterprise AI business case, and enforcing auditable pathways for sensitive information.

01

Clinical informatics / AI-assisted schema intelligence

Make a 600+ table clinical database easier to understand.

A schema assistant turns an unfamiliar 600+ table data model into an immediate, governed starting point for investigation and analyst enablement.

The problem

Troubleshooting an enterprise EMR across a multi-clinic network requires knowing how its data fits together. With more than 600 tables, learning the schema becomes part of every investigation and a major bottleneck for analysts.

What I built & enabled

I built an AI-assisted query and schema intelligence tool grounded in the clinical SQL Server data dictionary. Its scope is strictly metadata and schema: it guides investigation structure while enforcing air-tight data boundaries that keep patient records outside the assistant.

What changed

Analysts navigate complex relational schemas in minutes rather than hours, dramatically shortening root-cause investigation cycle times and accelerating onboarding across the clinical systems team.

  1. 01Clinical systems question
  2. 02Schema & metadata context
  3. 03AI-assisted query development
  4. 04Analyst validation
  • SQL Server (600+ Tables)
  • Schema Retrieval / RAG
  • Data Boundary Isolation
  • Analyst Enablement
  • Clinical Informatics
02

Clinical AI / Enterprise governance & routing architecture

Give clinical AI a controlled place to run.

The data boundary is an architectural mandate that every model, agent, and workflow must respect.

The problem

Deploying AI in healthcare requires knowing which tasks involve protected health information, which environments can receive it under an active BAA, and what scope each agent requires. Useful capabilities need an approved, governed path into production.

What I built & enabled

I authored the enterprise AI business case and established a BAA-governed Azure AI Foundry pathway. I architected a model routing gateway with role-scoped agents and traffic classification: sensitive workloads use BAA-covered or local inference; commercial model work is strictly isolated to metadata, code, and architecture.

What changed

Clinical and operational teams adopt AI with defined execution paths, auditable access boundaries, and zero risk of unauthorized PHI egress.

  1. 01Classify the work
  2. 02Check its data sensitivity
  3. 03Route to the permitted environment
  4. 04Scope agent access
  • Azure AI Foundry
  • Model Routing Gateway
  • BAA / PHI Governance
  • Enterprise AI Business Case
  • Local LLM Inference
Interactive Reference Architectures
Blueprint 01 / Multi-Agent RAG Hybrid Retrieval • Agent State Graph • Deterministic Guardrails

Enterprise-Scale Multi-Agent RAG & Validation Topology

Click any architectural plane below to inspect its mandate, stack patterns, and SLO telemetry—or run a live query trace across all four planes.

Live Query Tracer Idle / Ready
Tracing Query: "Analyze clinical protocol variance across multi-facility treatment records"
Plane 01: Validating schema & RBAC tags...
PLANE 01 DEEP-DIVE

Contextual Ingestion & Enrichment Plane

Throughput: 450 pgs/min
Architectural Mandate

Ingests raw enterprise and clinical documents (PDFs, protocol notes, regulatory filings, transcripts). Extracts hierarchical layout structure, preserves table headers, and injects parent-child lineage alongside role-based ACL metadata before indexing.

Under-the-Hood Patterns
Layout-Aware Parsing Unstructured Pipeline Dynamic Chunking (256-512) Redis Task Queue Parquet Lineage Store
Production Telemetry & SLO
Layout Parse F1:99.4%
ACL Metadata Tags:12 attributes
Index Sync Latency:< 1.2s
Blueprint 02 / Data-Lane Governance Workload routing by data sensitivity (BAA vs. Zero-PHI Lanes)
Clinical work schema / code / ops / patient records Data-lane classifier Does this touch PHI? LANE A / PHI PERMITTED Azure AI Foundry active BAA Local inference on-premise, no egress LANE B / NO PHI, EVER Commercial models schema / code / architecture Agent fleet scoped by role & lane Model routing gateway policy enforced in transit

02 / AI enablement & agentic workflows

Less repeated work. More organizational leverage.

AI enablement succeeds when complex capabilities become reliable, everyday workflows for clinicians, analysts, and operations staff. Select any initiative below to see how I combine weekly workflow discovery, reusable agent skills, persistent memory, and cross-model review gates.

03

AI enablement / Conversational workflow automation

Turn a conversation into structured enterprise intake.

Project, task, knowledge base, and change advisory board requests transform from repetitive form routines into reusable agent skills.

The problem

Clinicians and operations staff were manually creating tickets, filling out repetitive forms, and uploading attachments across disjointed intake paths. The administrative overhead drained focus from higher-value clinical priorities.

What I built & enabled

I built a library of reusable agent skills for project intake, task management, knowledge base documentation, and change advisory board (CAB) submissions. The skills connect conversational input directly to the structured record creation and upload APIs those processes require.

What changed

Routine intake begins with a natural conversation. Reusable agent skills automate record generation and document attachment, eliminating manual data entry and enabling consistent adoption across operational workflows.

  1. 01Describe the request
  2. 02Agent applies the intake skill
  3. 03Create the structured record
  4. 04Attach supporting material
  • AI Enablement
  • Reusable Agent Skills
  • Conversational Intake
  • Workflow Automation
  • Change Management
04

Agent systems / Persistent institutional knowledge

Stop rebuilding context every time work resumes.

Arm autonomous sessions with persistent memory, operational findings, and synthesized skills from previous runs.

The problem

Agents operating in isolated sessions repeatedly lose prior findings, procedural decisions, and runtime context. High-value operational discoveries remain trapped in ephemeral chat histories.

What I built & enabled

I built a shared, persistent knowledge base paired with an automated session review pipeline. The pipeline analyzes completed workflows, extracts repeatable operational patterns, and synthesizes them into modular skills available to the entire agent fleet.

What changed

Decisions and architectural discoveries compound across sessions. As agents resolve novel operational challenges, the platform automatically codifies those patterns into reusable enterprise skills.

  1. 01Complete a task
  2. 02Review the session
  3. 03Preserve findings & reusable steps
  4. 04Use them in the next task
  • Persistent Context
  • Multi-Agent Orchestration
  • Dynamic Skill Synthesis
  • Knowledge Retrieval
05

Software delivery / Cross-model evaluation gates

Build independent review into the agent workflow.

One frontier model implements the plan; an independent model family reviews and stress-tests it before completion.

The problem

When a single agent both generates and approves its own work, review often echoes the model's original blind spots and assumptions. Delivering production-grade code requires independent validation.

What I built & enabled

I built an agent collaboration and verification gateway where distinct model families share structured context, challenge architectural proposals, and hand off work across mandatory review gates.

What changed

Independent review is an automated delivery gate. Every implementation must pass cross-model scrutiny and deterministic criteria before work is marked complete.

  1. 01Shared task context
  2. 02Implementation by one model
  3. 03Review by another model
  4. 04Resolve findings before completion
  • Cross-Model Evaluation
  • Verification Gates
  • Multi-Model Consensus
  • Software Delivery
06

AI enablement / Organizational adoption & governance

Run weekly workflow discovery to turn friction into governed tools.

Sustainable AI enablement pairs executive business-case governance with a continuous, bottom-up cadence alongside clinicians and analysts.

The problem

Top-down AI initiatives often stall because staff either don't trust how tools handle sensitive clinical context or don't see how AI applies to their daily operational bottlenecks.

What I built & enabled

Alongside authoring the enterprise AI business case and BAA routing architecture, I lead weekly workflow discovery reviews with the clinical systems team. Together we inspect real clinical, administrative, and support bottlenecks, classify their data sensitivity, and turn the highest-friction tasks into governed assistants and reusable agent skills.

What changed

AI adoption became a structured, repeatable operational habit rather than a one-off experiment—steadily expanding automation coverage while keeping 100% compliance with clinical data boundaries.

  1. 01Weekly team workflow review
  2. 02Classify data & BAA boundary
  3. 03Build & test reusable skill
  4. 04Roll out to clinical/ops teams
  • AI Enablement Leadership
  • Workflow Discovery Cadence
  • Change Management
  • Clinician & Analyst Upskilling

Role-Aligned Capabilities Matrix

Built for clinical safety, enterprise architecture, and organizational adoption.

How my technical stack, healthcare domain leadership, and enablement practices align across AI Architect, Clinical AI, and AI Enablement roles.

AI Architecture & RAG Systems

  • Multi-agent orchestration & stateful supervisor graphs
  • Hybrid dense vector + sparse BM25 retrieval & RRF reranking
  • Model routing gateways (Azure AI Foundry, OpenAI, Claude, Local)
  • Cross-model verification gates & deterministic faithfulness evals
  • Persistent session memory & dynamic skill synthesis pipelines
  • Python, .NET 8, React/TypeScript, SQL Server, REST APIs, Git

Clinical AI & Data Governance

  • HIPAA / BAA data-lane boundary & zero-PHI egress architecture
  • 600+ table enterprise EMR SQL Server schema RAG & intelligence
  • Multi-EMR clinical data consolidation (3M+ records, Entra ID SSO)
  • Inline PII/PHI redaction filters & OpenTelemetry audit lineage
  • Emergency nursing (RN) & perioperative clinical workflow design
  • HL7 / FHIR interoperability & row-level authorization controls

AI Enablement & Transformation

  • Enterprise AI business cases, governance playbooks & roadmaps
  • Weekly clinical & operational workflow automation discovery reviews
  • Reusable agent skill libraries for intake, CAB, tasks & documentation
  • Translating bedside clinician & analyst friction into governed tools
  • Multimillion-dollar vendor portfolio & SLA lifecycle leadership
  • 95% on-time delivery across enterprise upgrades & cloud migrations

03 / Supporting engineering

Problems solved in clinical systems.

My experience shows up in the problems I investigate and the tools I build. These examples from my clinical systems work connect the underlying fault to a practical solution.

Clinical record access

Bring four legacy EMRs into one view.

Problem: Records were spread across four legacy platforms, with shared credentials for access.

Solution: I built a read-only clinical viewer and migration tooling with .NET, React, and Azure. The migration covered 3M+ clinical records, 770K+ documents, and 80K+ treatment cycles. Entra ID single sign-on replaced shared credentials, with authorization controls across clinics.

Safer operational changes

Check the system before applying the plan.

Problem: The live system could change between approval and execution.

Solution: My Python/Flask administration console handles chart merges, provisioning, drug libraries, and permissions. It re-validates live state before execution and stops when conditions differ from the approved plan. It caught a real divergence on its first run.

Data integrity

Find the fault behind a successful load.

Problem: A laboratory migration reported success while a non-unique key collapsed distinct records.

Solution: I traced the fault, changed the keying approach, and recovered records that would otherwise have been lost.

EMR troubleshooting

Find charts the system made invisible.

Problem: A vendor function made affected charts impossible to open and invisible in search, making missing records harder to recognize.

Solution: I traced the issue to a null-dereference and produced a reversible fix.

Permission governance

Verify the audit trail behind access changes.

Problem: During access remediation across two production sites, partial permission changes were missing from the vendor’s audit trail.

Solution: I identified the stored-procedure gap, removed excess access, and validated the governed change.

Patient document access

Put access controls around sensitive documents.

Problem: A patient document store allowed public reads, and a separate authorization gap permitted access across clinics.

Solution: I implemented authenticated document delivery, row-level controls, and a global authorization filter.

04 / The person behind the work

Clinical experience shapes what I build.

Duke Morgan

I started in emergency nursing, where a difficult workflow has immediate consequences for the person trying to use it. That experience still shapes the questions I ask: What is someone trying to do? Where does the system get in the way? How will we know the change worked?

Today, I direct clinical systems at Inception Fertility and continue to build software and AI tools myself. I bring the clinical context, implementation experience, and technical depth needed to take a problem from investigation through delivery.

My interests extend into applied AI and automation wherever people need help navigating complex systems and repetitive work.

2005 – 2013 • Bedside Clinical Foundation Emergency Department Charge Nurse (RN) HCA Southern Hills & Williamson Medical Center Led high-acuity ED nursing operations and served as CPOE superuser bridging bedside care and IT.
2013 – 2017 • Enterprise EMR Rollouts Clinical Implementation & Corporate Analyst MEDHOST & Community Health Systems (CHS) Designed workflows and led multi-hospital EMR go-lives across emergency, OR, PACU, and med-surg.
2017 – Present • Executive & AI Architecture Director of Clinical Systems & Applied AI Multi-State Fertility Care Network Direct multi-EMR strategy, $MM vendor portfolio, BAA-governed Azure AI Foundry routing, and agentic tools.

05 / Verified credentials & foundation

Clinical licensure meets frontier AI.

Active multistate clinical licensure and health administration leadership, backed by 12+ verified AI and data architecture credentials across OpenAI, Anthropic, Microsoft, and Google. Select any category to inspect and verify individual credentials.

Clinical Licensure, Healthcare Leadership & Data Architecture (4)
Active RN Multistate Compact BS Healthcare Administration (WGU) Microsoft SQL Server Specialization AAS Nursing
Active

Registered Nurse (RN)

Tennessee Board of Nursing  /  Multistate Compact

Active Tennessee / multistate compact license. Clinical foundation in emergency care and clinical systems intuition where safety and workflow accuracy carry real-world consequences.

Specialization  /  2026

Microsoft SQL Server Specialization

Microsoft  /  Coursera Specialization Certificate

Five-course specialization spanning T-SQL, relational schema design, indexing, performance optimization, and database security across enterprise clinical data models.

Verify credential
Degree  /  2025

BS, Healthcare Administration

Western Governors University

Health systems, healthcare finance, regulatory compliance, and operations, completed while directing clinical systems full time.

Degree  /  2007

AAS, Nursing

Excelsior College

Nursing education supporting a foundational career in emergency bedside care and healthcare technology leadership.

AI Architecture, Agent Engineering & Performance (6)
OpenAI: Optimize AI Performance OpenAI: Agents & Workflows Microsoft: Azure AI Foundry Agents Claude: AI Fluency for Builders OpenAI: AI Foundations Claude: Framework & Foundations
Credential  /  2026

Optimize AI Application Performance

OpenAI Academy  /  Accredible

Using production evidence to evaluate AI system behavior under real traffic: comparing model versions, spotting regressions, and engineering deliberate trade-offs across quality, latency, throughput, reliability, and cost.

Verify credential
Credential  /  2026

Agents and Workflows

OpenAI Academy  /  Accredible

Directing autonomous agents through structured enterprise work: context engineering, output boundaries, iterative draft reviews, workflow synthesis, and reusable agent skills.

Verify credential
Course  /  2026

AI Fluency for Builders

Claude Academy  /  Anthropic

Hands-on system building, prompt engineering architecture, API integration patterns, and builder workflows using Claude models.

Verify credential
Course  /  2026

AI Agent Fundamentals with Azure AI Foundry

Microsoft  /  Coursera Course Certificate

Agent design, orchestration, and Azure AI service integration, operating on the same BAA-governed platform used for production clinical AI routing.

Verify credential
Credential  /  2026

AI Foundations

OpenAI Academy  /  Accredible

Core LLM architecture and foundations, prompt engineering with rich context, output review and evaluation, and responsible enterprise AI practices.

Verify credential
Course  /  2026

AI Fluency: Framework & Foundations

Claude Academy  /  Anthropic

Comprehensive foundation in large language model capabilities, frontier model safety guardrails, systemic risk evaluation, and effective AI deployment.

Verify credential
AI Enablement, Strategy & Organizational Leadership (6)
OpenAI: AI Leadership OpenAI: Scope AI Solutions Google: Generative AI Leader Specialization Claude: Small Business Claude: Nonprofits Claude: Creative Work
Credential  /  2026

AI Leadership

OpenAI Academy  /  Accredible

Leading organizational AI adoption and transformation: developing sequenced AI roadmaps around measurable value, establishing governance and sponsorship, and authoring executive AI Strategy Briefs.

Verify credential
Credential  /  2026

Scope AI Solutions

OpenAI Academy  /  Accredible

Identifying high-leverage AI opportunities inside operational workflows, evaluating business value against implementation requirements, and framing technical solution plans.

Verify credential
Specialization  /  2026

Google Generative AI Leader

Google  /  Coursera Specialization Certificate

Enterprise generative AI strategy, business model innovation, and organizational AI transformation frameworks completed in 2026.

Verify credential
Course  /  2026

AI Fluency for Small Businesses

Claude Academy  /  Anthropic

Practical AI adoption, operational efficiency, customer workflow automation, and high-ROI implementation strategies for lean teams.

Verify credential
Course  /  2026

AI Fluency for Nonprofits

Claude Academy  /  Anthropic

Operational leverage, resource optimization, and workflow automation strategies tailored to mission-driven and resource-constrained environments.

Verify credential
Course  /  2026

AI Fluency for Creative Work

Claude Academy  /  Anthropic

Generative synthesis, multi-step prompt design, iterative ideation, and collaborative problem framing with frontier models.

Verify credential

Have a problem worth solving?

Let's talk about how clinical intuition, governed AI architecture, and operational enablement can accelerate your organization.

AI Architect Clinical AI Leadership AI Enablement & Transformation Director / VP of Clinical Systems
View Résumé (PDF) Download DOCX Connect on LinkedIn Email Duke