Customers

Teams that run AI where the work happens

Utilities, defense contractors, and hospital networks use Gammaxon to put models in places a hosted service cannot reach. Three recent engagements, in the customers' words and numbers.

Customer stories

Energy and utilities

Pinecrest Utilities

Who: Regional electric utility, 2.1 million meters
Deployment: On-premises, two control centers
Time to production: 10 weeks

The problem

Control-room operators spent hours of every shift reading alarm logs and work orders across five systems. The security team would not let operational data leave the utility's network, which ruled out every hosted AI service they evaluated.

What was deployed

Gammaxon Workbench on two on-premises clusters, one per control center, with an alarm-summarization model tuned on the utility's own history. Our engineers worked in the control room for ten weeks and trained the utility's team to operate it.

Result: alarm triage time down 41% in the first quarter, with no operational data leaving the utility's network.

Aerospace and defense

Kestrane Defense Systems

Who: Defense contractor, six facilities
Deployment: Fully air-gapped enclave
Time to production: 7 weeks

The problem

Engineers needed to search and summarize decades of design documents inside an enclave with no outside connection, and every software update had to pass a media-transfer review before it crossed the boundary.

What was deployed

An air-gapped deployment kit installed from signed media, a document-search workspace, and a catalog of three approved models. Updates arrive as signed bundles that the customer's security office verifies before they enter the enclave.

Result: first production workload in seven weeks, and the first signed update applied three hours after it arrived on site.

Healthcare

Tessellate Health

Who: Hospital network, 9 hospitals and 140 clinics
Deployment: Private data center, patient-data zone
Time to production: 9 weeks

The problem

Clinical documentation took hours of physician time every week, and the network's privacy policy did not allow patient data to be sent to an outside model provider.

What was deployed

A documentation-assistant workspace inside the network's own data center, a model catalog gated by the clinical informatics committee, and a partner API that the records system calls over the internal network.

Result: physicians regained an average of 5.5 hours a week, and the privacy office approved the deployment in a single review cycle.

In their words

We have turned down every AI vendor that asked us to open a port. Gammaxon asked for a loading dock and a signed bundle, then taught our own people to run it.

Teodor Lindqvist
Director of Systems Security, Kestrane Defense Systems

Tell us where your model has to run

An engineer will tell you how long it takes to get there, and what we would need from your team.

Talk to an engineer Explore the platform