AI Agents Migrate Healthcare Platform from AWS to Google Cloud

Details

This case study addresses the challenges faced by a multi-tenant healthcare interoperability platform built on AWS (supporting FHIR R4). The system ingests and normalizes clinical data, requiring migration to Google Cloud while maintaining HIPAA compliance, multi-tenant isolation, and core FHIR/HL7 interoperability. The sheer size and complexity of the codebase made manual porting unfeasible. Furthermore, the move required significant architectural decisions on GCP—including eventing backbone design (Kafka vs. Pub/Sub), identity/OAuth2 model, and multi-tenant topology—beyond a simple lift-and-shift. Leopard Data co-built an AI migration agent in TypeScript. This agent operates as a multi-pass pipeline: it generates documentation from the AWS source, feeds that context back into the model, and then produces the ported GCP implementation. Architecturally, they mapped services (e.g., HealthLake to Google Healthcare API; DynamoDB to Firestore) and resolved complex issues like establishing a centralized Pub/Sub + EventArc eventing backbone and building a vendor-agnostic abstraction layer for messaging (supporting Kafka optionality). This approach successfully enabled the migration of a large, regulated healthcare system across cloud boundaries.