Member of Technical Staff, Performance Engineering

Physical Superintelligence is a startup with roots at Google, NVIDIA, Harvard, Meta, MIT, Oxford, Johns Hopkins, Cambridge, and the Perimeter Institute building AI systems to discover new physics at scale. We are seeking a performance engineer who owns a workload's time end to end, from the application code and the data path through the scheduler to the GPU kernels, across AI and HPC, to make one cluster produce more physics.

Our mission is to discover and commercialize transformative physics breakthroughs at scale with artificial superintelligence, safely, verifiably, and for broad public benefit.

The last century's golden age of physics gave us transistors, lasers, and nuclear energy. We believe artificial superintelligence will unlock the next one. We're creating the infrastructure to industrialize scientific discovery and usher in this new era.

We have one product: new physics, at scale.

• Own a workload's time end to end: when a training loop, a simulation or a served model runs slow, find the layer and fix it there, in the owner's repository, with a benchmark left behind.
• Own the kernels: specify, generate or write, verify and profile the GPU code physics campaigns and learned models need.
• Go deep in one lane: multi-node AI training, inference serving at scale, or HPC simulation at scale.
• Measure the cluster on real jobs so that queue policy and the next capacity decision rest on numbers somebody observed.

• Three or more years of whole-stack performance work. You have taken a workload from slow to fast across more than one layer, with a trace that shows where the time went before and after, and at least one of those layers was not the GPU: a data loader, a filesystem, a scheduler, a container start, a service call, a Python hot loop. You work from measurements, instrument systems you did not write, and verify a win on a real workload before you claim it.
• Three or more years of systems programming. You read, modify and verify C++, CUDA or Triton code that someone or something else wrote, and say whether it is correct. You work from profilers (Nsight Systems and Compute, torch profiler, perf, NCCL tests) and reason about memory hierarchy, occupancy and the roofline.
• Scale in one of the following, and the appetite for the others. The cluster runs all three; you bring one and learn the rest on our hardware.
• Multi-node AI training: distributed training throughput, collective-communication behavior, InfiniBand topology, RL sampler and trainer placement, where step time goes and why MFU disappoints.
• Inference serving at scale: vLLM, SGLang or comparable, multi-GPU placement of mixture-of-experts models, KV-cache and quantization trade-offs, what sets the latency and throughput floor on a fixed set of GPUs.
• HPC simulation at scale: a CFD, molecular dynamics, multiphysics or comparable code taken across nodes with MPI, scaling measured, solver kernels ported or tuned on GPUs, FP64 and extended-precision numerics.

• Kernels you wrote yourself in CUDA C++ or Triton that other people's systems depend on.
• A second lane from the list above.
• Source-level work in OpenFOAM, LAMMPS, GROMACS, or comparable codes; PETSc, Trilinos, MAGMA or comparable libraries.
• Kueue or Slurm queue policy on a shared cluster, with the before-and-after that justified a change.
• Experience turning research code into a maintained library that a team of scientists depends on.

Work Authorization: Candidates must be authorized to work in the United States. PSI is not able to sponsor or take over sponsorship of employment visas for this role, now or in the future.

We hold a high technical bar and give people full ownership of their work, from spec to ship to on-call. We write contracts before logic, test against real systems instead of mocks, and favor simple designs that ship over clever ones that do not. Our development process is AI-native: we work with agentic coding tools daily, write specs that are legible to humans and agents alike, and lead with leverage.