Millrace

Millrace operates hospitals. Notes cannot leave the network, and a clinician will not wait on a search that feels like a report.

Hybrid Cloud keeps the data plane on their Kubernetes clusters. Queries filter by facility and note type while scoring.

In this story, median retrieval of a relevant passage stays under 40 milliseconds on the ward network.

The model that writes the summary runs beside the index. The vector database is not the system of record for the chart.