Teleconsultation Platform Migration: SFTP to Real-Time API

Swati inherited the telemedicine product charter at a point where Amazon Pharmacy's teleconsultation system was running on an SFTP-based architecture. Every request sent to a third-party telemedicine partner sat in an S3 bucket until the partner pulled it, and the reverse leg worked the same way. The round-trip lag was upwards of 20 minutes, meaning customers who had just placed an order were waiting that long before a doctor called them.

The first task was to determine whether any existing internal solution could solve the problem. Swati analyzed alternatives across Amazon's internal ecosystem, including a service called Iris used by Amazon Clinic, and presented the findings to leadership. The conclusion was clear: none of the existing solutions could handle external endpoints, multi-partner support, and the health data security requirements without a rebuild that would take as long as building new.

With that case made, she led the design and build of a new API-native platform. The new architecture enabled real-time request creation and response consumption, and unlocked a capability that had not existed before: order cancellation propagation. Previously, if a customer cancelled an order after a teleconsultation request had been sent, there was no way to stop the partner from continuing to call. The new platform closed that loop.

On top of the core migration, Swati built granular call-status signals into the system, so that each time a partner attempted to reach a customer, Amazon received a real-time update. She is now building notification nudges on top of those signals, prompting customers who have missed calls to request an immediate callback.