arun.v

Healthcare SaaS · Agentic AI · Mar 2025 to Sep 2025

EMA Clinical Workflows & mmRxService: Electronic Prescribing

Python/FastAPI e-prescribing on AWS with Bedrock-backed GenAI clinical decision support and a LangGraph prior-authorization workflow.

Senior Software Engineer
HealthcareEnterprise SaaSAI / ML
CDS alerts
Bedrock-backed, inline, real-time
PA drafting
LangGraph agent, human-in-the-loop
Identity
Okta RBAC, DEA EPCS / NPI-compliant

Problem and Approach

Electronic prescribing modernization at a clinical SaaS platform: build the clinical user interface and the AWS-hosted backend pipeline for sending prescriptions electronically to pharmacies, then layer Generative AI onto the workflow without letting free-form model output into a regulated clinical path.

  • GenAI clinical decision support. A drug interaction alert panel backed by Amazon Bedrock surfaces real-time LLM-generated clinical guidance inline in the prescribing workflow. Guidance is presented to the clinician; it never auto-acts.
  • Agentic prior authorization. A LangGraph-based workflow drafts PA requests autonomously from clinical context, routes every draft to a human reviewer for approval, and submits to payers with audit-logged escalation paths. Bedrock Agents provide the agentic layer behind CDS and PA drafting with human-in-the-loop review gates.
  • Event-driven prescription processing. Prescription orders flow through SQS/SNS and Lambda, with Redis for low-latency session state and PostgreSQL for prescription records. The event-driven shape keeps the prescription path resilient to transient SureScripts failures and rate-limit windows.
  • Contract-first APIs. Prescribing and medication-history endpoints documented with Swagger/OpenAPI, consumed by downstream pharmacy and payer partners.
  • Identity and authorization. Okta-based RBAC scoped to DEA EPCS and NPI requirements, ensuring only credentialed providers submit controlled-substance prescriptions.
  • Architectural carryover. NemoRx SureScripts and FHIR API patterns (retry semantics, partial-write recovery, expired-token handling) carried directly into mmRxService, shortening team ramp-up on the e-prescribing surface. Clinical workflow expertise from prior podiatry EHR work guided charting, billing, and scheduling decisions before they reached implementation.

Outcomes

  • Python/FastAPI prescribing backend in production with the Bedrock-backed CDS drug interaction panel inline in the clinical workflow.
  • LangGraph prior-authorization drafting workflow behind a human review gate, with audit-logged escalation paths to payers.
  • Event-driven pipeline on AWS SQS/SNS and Lambda resilient to transient SureScripts failures.
  • DEA EPCS / NPI-compliant RBAC on Okta delivered alongside the prescription flow.

Stack

Python 3.11FastAPILangGraphAmazon BedrockAmazon Bedrock AgentsAmazon SageMakerAWS LambdaAWS SQS / SNSRedisPostgreSQLAWS IAMSwagger / OpenAPIOktaSureScriptsFHIR / SMART on FHIRICD-10 / CPTNPI / DEA / EPCS

Highlights

  • Built the Python/FastAPI backend for the clinical prescribing and medication-history workflow, embedding a Generative AI clinical decision support (CDS) drug interaction alert panel backed by Amazon Bedrock that surfaces real-time LLM-generated clinical guidance inline.
  • Designed a LangGraph-based agentic prior-authorization workflow that autonomously drafts PA requests from clinical context, routes to a human reviewer for approval, and submits to payers with audit-logged escalation paths.
  • Built event-driven prescription processing on AWS with SQS/SNS and Lambda, Redis for low-latency session state, and PostgreSQL for prescription records, resilient to transient SureScripts failures.
  • Applied NemoRx SureScripts and FHIR API architecture patterns to accelerate team ramp-up on mmRxService, and advised on DEA EPCS / NPI-compliant RBAC with Okta in a regulated pharmacy workflow.