James Youm

AIScribe - AI-Powered Telehealth Documentation Platform

Multi-Model AI Pipeline - Automated SOAP Note Generation with HL7 v2 EHR Integration

I designed and built AIScribe as a full-stack engineering showcase to bridge my clinical background with enterprise healthcare integration. The primary purpose of this project was to dive deep into the technologies that power modern hospital interoperability — specifically Mirth Connect, HL7 messaging standards, and FHIR APIs — while exploring how to safely introduce AI transcription into a clinical data pipeline. AIScribe simulates a real-world integration loop: it takes a telehealth provider's raw audio session, generates a structured clinical SOAP note using multi-model AI, and securely routes the validated document across our network perimeter into an isolated, simulated EMR database.

• Multi-Model AI Transcription Pipeline: Engineered an end-to-end pipeline using OpenAI Whisper for transcription and Anthropic Claude for structured SOAP note generation, enforced by a three-step JSON schema validation layer.
• HL7 v2 EHR Integration: Integrated Mirth Connect as a clinical messaging broker to transform application JSON into structurally compliant HL7 MDM^T02 and MDM^T04 messages, routing them securely to a simulated EMR database.
• FHIR R4 Schedule Sync: Implemented a mock FHIR R4 gateway that translates raw EMR database records into compliant FHIR Bundle resources to sync provider schedules into a local application cache.
• Immutable Note Versioning: Developed an EMR amendment workflow where signed clinical records are never modified in-place, utilizing a versioned database schema with an "is_current" flag pattern managed through Mirth.
• PHI Security & Compliance: Implemented transient, in-memory audio processing that avoids disk writes, enforced provider-scoped SQL isolation to prevent multi-tenant data leaks, and built asynchronous HIPAA audit logging.
• Idempotent Note Submission: Built a client-server idempotency guard utilizing UUID4 client-side keys and backend database uniqueness constraints to eliminate duplicate EMR writes from network drops or rapid double-clicks.

Technical Stack: React 18, Vite 5, Node.js 18, Express 4, PostgreSQL 15, OpenAI Whisper API, Anthropic Claude API, Mirth Connect 4.x, HL7 v2, FHIR R4, JWT, bcrypt, Docker Compose

View Project on GitHub

Real-Time EMS Handoff Application (Full-Stack Cloud Project)

Azure Serverless Pipeline — Real-Time EMS-to-ED Clinical Data Streaming

EMS crews routinely transport critically ill patients to Emergency Departments with no structured channel for clinical data to travel ahead of them. ED staff have no advance visibility into vitals, administered medications, ESI triage level, or required resources until the gurney arrives — forcing a reactive workflow that creates intake bottlenecks and delays resource activation. To close this gap, I designed and built an end-to-end serverless data pipeline that streams pre-hospital patient data from EMS field providers to clinical ED dashboards in real time. The hospital dashboard consolidates incoming patient data into clear, actionable insights, giving clinical leadership the visibility needed to manage staffing and resources before EMS arrival.

• Type-Safe Frontends: Developed responsive Progressive Web Apps (PWAs) using React 18 and TypeScript (Vite) for both the EMS-facing patient ingestion application and the hospital-facing clinical dashboard.
• Data Validation & PHI Ingestion: Engineered a backend ingestion engine using Azure Functions and Pydantic to enforce strict schema validation on incoming payloads containing PHI prior to database persistence.
• Event-Driven Database Pipeline: Configured Azure Cosmos DB as the primary data store, leveraging the Cosmos DB Change Feed to instantly trigger serverless event handlers the moment an EMS payload is successfully written.
• Real-Time WebSocket Streaming: Integrated Azure SignalR to capture change feed events and broadcast live, low-latency updates over WebSockets directly to the hospital-facing dashboard PWA, removing any need for manual data fetching.

Technical Stack: React 18, TypeScript, Vite, Python, Pydantic, Azure Cosmos DB, Azure Functions, Azure SignalR, Azure Blob Storage, Azure Static Web Apps, HL7 FHIR

View Project on GitHub | EMS Medic PWA | Hospital Dashboard

Live demo requires Entra ID login — credentials available on request.

Enterprise Healthcare Interoperability Middleware Design (WGU Capstone)

Multi-Cloud Architecture Design — EMS-to-ED Platform at Enterprise Scale

Designed and documented a production-scale enterprise architecture extending the EMS Handoff App concept from a single-agency pilot to a multi-tenant, multi-cloud healthcare interoperability platform. This capstone project focuses on the architectural design decisions required to run shared healthcare infrastructure at city scale — multiple EMS agencies, multiple hospital systems, and strict per-tenant PHI data isolation — rather than implementation of a deployed system.

• Pub/Sub Routing Architecture: Designed a multi-tenant Azure Service Bus topology with per-hospital topics and per-ED subscriptions using SQL filter rules, routing NEMSIS payloads to the correct hospital system and emergency department without cross-tenant data exposure.
• Zero Trust Security Posture: Architected a private-by-default Azure VNet topology with private endpoints on all PaaS services, system-assigned Managed Identities with least-privilege RBAC, customer-managed key (CMK) encryption at rest via Azure Key Vault, and Azure Policy guardrails enforced at the management group level.
• Cross-Cloud Disaster Recovery: Designed a three-tier AWS warm standby architecture — always-active S3 Object Lock HIPAA compliance archive, Route 53 DNS health-check failover (~2.5 min RTO) to a pre-provisioned AWS Lambda/SQS/HealthLake pipeline, and a Phase 2 dashboard continuity layer using AWS Amplify and Amazon Cognito.
• FHIR ETL Pipeline Design: Specified a NEMSIS v3.5.0 → FHIR R4 transformation engine with a per-hospital schema registry pattern, composite idempotency key enforcement, and HL7 R4 conformance validation gating before any FHIR resource reaches the store.

Architectural Components (System Design): Azure Service Bus, Azure API Management, Azure Functions, Azure Health Data Services (FHIR R4), Azure Cosmos DB, Azure Key Vault, Azure SignalR, Microsoft Sentinel, AWS S3, AWS Lambda, Amazon SNS/SQS, AWS HealthLake, Amazon Route 53, HL7 FHIR R4, NEMSIS v3.5.0

View Architecture Documentation on GitHub

Personal Cloud Resume Website Deployment

AWS Serverless Architecture — Static Frontend, Lambda Backend, and CI/CD Pipeline

Demonstrates full-stack cloud proficiency by hosting a static HTML/CSS/JavaScript frontend on AWS S3, delivered globally via CloudFront and secured with Route 53/ACM. A serverless AWS Lambda function (Python), exposed via Amazon API Gateway, leverages DynamoDB for a dynamic visitor counter. The architecture is fully automated with GitHub Actions CI/CD pipelines for both frontend and backend deployments.

Technical Stack: HTML, CSS, JavaScript, Amazon S3, Amazon CloudFront, Amazon Route 53, AWS ACM, AWS Lambda (Python), Amazon API Gateway, Amazon DynamoDB, AWS IAM, AWS SAM, GitHub Actions

View Project on GitHub

Azure Resource Janitor Script

PowerShell Cloud Governance — Automated Orphan Detection and Cost Optimization

Developed a PowerShell automation suite to maintain cloud hygiene and reduce costs by identifying and decommissioning orphaned or expired Azure resources — including unattached managed disks and resource groups that have exceeded their expiration tag. Built with a dry-run toggle to safely preview all deletions before execution and tag-based protection to ensure critical resources are never accidentally removed.

Technical Stack: PowerShell, Az PowerShell Module, Azure Resource Manager, Git, GitHub

View Project on GitHub

PowerShell Automation Scripts for IT Operations

Windows Server Automation — Active Directory, SQL Server, and System Administration

Developed a collection of PowerShell scripts to automate common IT administrative workflows. Built an interactive, menu-driven system administration script for file management and real-time server monitoring (CPU/memory usage, process inspection). Additionally engineered targeted infrastructure automation scripts: an Active Directory script that manages OU lifecycle and bulk-provisions users from a CSV file, and a SQL Server script that automates complete database restoration, table creation, and data import.

Technical Stack: PowerShell, Active Directory, SQL Server, Windows Server Administration

View Project on GitHub