Give every AI agent its own GPU Senior Software Engineer – GPU Sandboxes
San Francisco onsite | Relocation considered | Equity offered
AI agents increasingly need more than access to a model. They need their own secure compute environment, with real GPU capacity available in seconds.
That is what you’ll help build.
You’ll join a well-funded AI infrastructure company creating isolated, stateful computers that agents can launch and control through an API. The GPU platform is still being built, so you won’t be maintaining someone else’s architecture you’ll be instrumental in defining how it works.
You’ll work deep in the stack across:
KVM/QEMU and GPU passthrough
VFIO, IOMMU and PCIe isolation
NVIDIA drivers and CUDA lifecycle
VM boot paths, snapshots and cold-start performance
GPU-aware scheduling and bin-packing
Health monitoring, failure detection and automated recovery
Security boundaries across shared hosts
They need someone who has built or debugged the underlying virtualization, isolation or device layer in production. KVM/QEMU, Cloud Hypervisor or Firecracker would be ideal, although similarly deep experience with crosvm, Kata Containers, Xen or another hypervisor can translate.
You should also have operated GPUs in production and be comfortable tracing difficult failures across the hypervisor, Linux kernel, drivers and physical hardware.
Go, Rust or C/C++ can all work.
This is rare, high-impact systems engineering with very little existing playbook. You’ll have genuine influence over the architecture, technical direction and foundations of a GPU platform being built for the next generation of AI agents.
Onsite in San Francisco, with relocation considered and equity offered.
#J-18808-Ljbffr