Hello There!

Welcome to our platform.

ABOUT CALADRIUS

We find what breaks
in GPU infrastructure, and we fix it.

Caladrius is root-cause analysis and closed-loop remediation for GPU infrastructure. When a job stalls or slows, it traces the problem below the application layer, across device, fabric, storage, and workload, then drives the fix on your approval and verifies it held.

WHO WE ARE

GPU clusters degrade before they fail, and both happen beneath the application layer. Keeping them reliable is an infrastructure-software problem: distributed systems, automation, operations at scale. We have built that software before. The team founded ReleaseIQ, an SRE and ops-automation platform acquired by CloudBees, after years at VMware and WebLogic.

OUR MISSION

GPU infrastructure fails in ways general tooling can't see. A slowdown or failure in one layer can silently stall thousands of GPUs, with the root cause buried deep in the stack, and resolving it today costs hours of hand-correlation. Our mission is to close that gap: name the root cause, drive the fix, and verify the outcome, after a failure or ahead of one.

OUR TEAM
Seetharam Param
Seetharam Param
Co-founder & CEO
CEO / co-founder, ReleaseIQ (acq. CloudBees). VMware, WebLogic.
Sudhish Mangalasary
Sudhish Mangalasary
Head of Engineering
Director, CloudBees. Head of Eng, ReleaseIQ. Deputy GM, HCL America.
Chidambara Rajan
Chidambara Rajan
Co-founder, India Ops
Founder & MD, Awan Infotech (SRE services).
Marc Leavitt
Marc Leavitt
Head of Marketing
Marketing leadership at Iron Mountain, Brocade, EMC, Western Digital.
OUR ADVISORS
Sandhya Sridharan
Sandhya Sridharan
JPMorgan, VMware.
Scott Hammond
Scott Hammond
Node.js Foundation, CNCF, Cisco.
Siddanagouda Sankanagouda
Siddanagouda Sankanagouda
Sr. Director, Intel; co-founder, ReleaseIQ.
HOW WE BUILD

Four promises

Root cause, not symptoms

Restarts and reroutes keep a job limping. We go after the actual cause, so the same failure doesn't come back tomorrow.

No fix without proof

Nothing changes in your cluster without your approval, and every change is verified against what the system actually did. If it didn't hold, you know.

Native to GPU infrastructure

Device, fabric, scheduler, training, and serving failure modes are our home ground, not an integration afterthought.

Ahead of failure

A precursor pattern gets the full loop, not just an alert: the same diagnosis, fix, and verification, before the job goes down. The cheapest incident is the one that never opens.