Member of Technical Staff, Sandbox Infrastructure

San Francisco Tensor Company•San Francisco, CA

17h•Onsite

About The Position

SF Tensor is building high-performance compute infrastructure for AI, including a Kernel Optimizer and Model Foundry. The company is developing a serverless GPU container service that can run untrusted, freshly generated kernels and other workloads across NVIDIA, AMD, TPU, and Trainium accelerators. These sandboxes support compiler benchmarking and search, the company’s post-training runs, and isolated customer workloads. The role focuses on expanding and improving the sandbox infrastructure: adding vendor and accelerator support, building GPU virtualization below the runtime, improving measurement and profiling fidelity, reducing cold starts, and enabling reliable operation on preemptible capacity. Sandboxes may use multiple GPUs or nodes and need to handle failures, preemption, checkpointing, rescheduling, and live migration while preserving active sockets. The role works closely with compiler, post-training, and kernel teams to ensure sandbox performance does not limit their throughput. SF Tensor is a small team working on systems at frontier scale. Most work takes place in person at the San Francisco office. Relocation assistance is offered, and the company would like employees to work from the office as often as possible. This is a full-time position with a base salary range of $275,000–$315,000, plus equity and benefits.

Requirements

Strong low-level systems engineering background, including Linux kernel internals, containers, namespaces, cgroups, syscall interception, or hypervisors.
Experience in GPU systems engineering, such as drivers, runtimes, or scheduling on accelerator fleets.
Comfort with distributed-systems failure modes, including preemption, partial failure, checkpoint/restore, and handling state that cannot be lost.
Proficiency in Go, C/C++, or Rust.
Strong bias toward building solutions when vendors do not support the required capabilities.

Nice To Haves

Direct experience with gVisor, Firecracker, Kata, QEMU/KVM, or similar systems.
Direct experience with CRIU, live migration, or connection-preserving failover.
Familiarity with NCCL/RCCL, RDMA, InfiniBand, or vendor interconnects.
Experience running large fleets on spot or other preemptible capacity.
Familiarity with bare-metal provisioning, hypervisors, or fleet management at scale.
Security background in isolation boundaries and untrusted code execution.

Responsibilities

Extend the sandboxing stack to new vendors and accelerators.
Build and maintain GPU virtualization below the runtime, including gVisor work at the driver and ioctl level.
Make sandboxes first-class citizens on spot capacity through preemption-aware scheduling, checkpointing, and rescheduling.
Support multi-GPU and multi-node sandboxes, including the required NVLink, NVSwitch, InfiniBand, and RoCE paths.
Own live migration end to end, including socket-preserving migration.
Ensure measurement and profiling fidelity and isolation.
Work with compiler, post-training, and kernel teams to ensure sandbox infrastructure does not cap their throughput.