Member of Technical Staff, Performance Engineering

Physical Superintelligence•Boston, MA
•Hybrid

About The Position

Physical Superintelligence is a startup building AI systems to discover new physics at scale. Its mission is to discover and commercialize transformative physics breakthroughs using artificial superintelligence, safely, verifiably, and for broad public benefit. The company is creating infrastructure to industrialize scientific discovery. The Performance Engineer owns workload performance end to end, from application code and data paths through schedulers to GPU kernels, across AI and HPC. The role focuses on identifying and fixing performance bottlenecks in the appropriate codebase, developing and profiling GPU kernels, specializing in at least one of multi-node AI training, large-scale inference serving, or HPC simulation, and measuring real cluster workloads to inform queue policy and capacity decisions. The role is hybrid, with three days a week in person at the Cambridge, Massachusetts office. Candidates must be authorized to work in the United States; the company cannot sponsor or take over sponsorship of employment visas. The work may be subject to U.S. export control laws, which may require the candidate to qualify as a U.S. person or be eligible for required government authorization. The company is an equal opportunity employer.

Requirements

  • At least three years of whole-stack performance engineering experience, including taking a workload from slow to fast across multiple layers and using traces to show where time was spent before and after.
  • At least one performance layer outside the GPU, such as a data loader, filesystem, scheduler, container start, service call, or Python hot loop.
  • Ability to work from measurements, instrument systems not personally written, and verify performance gains on real workloads.
  • At least three years of systems programming experience.
  • Ability to read, modify, and verify C++, CUDA, or Triton code written by others, and assess its correctness.
  • Experience with profilers such as Nsight Systems and Compute, torch profiler, perf, or NCCL tests, and ability to reason about memory hierarchy, occupancy, and the roofline.
  • Experience at scale in at least one of multi-node AI training, inference serving, or HPC simulation.
  • Authorization to work in the United States; the employer does not sponsor employment visas.

Nice To Haves

  • Kernels written in CUDA C++ or Triton that other people's systems depend on.
  • Experience in a second area among multi-node AI training, inference serving at scale, and HPC simulation at scale.
  • Source-level experience with OpenFOAM, LAMMPS, GROMACS, or comparable codes, or PETSc, Trilinos, MAGMA, or comparable libraries.
  • Experience with Kueue or Slurm queue policy on a shared cluster, including before-and-after evidence justifying a change.
  • Experience turning research code into a maintained library used by a team of scientists.

Responsibilities

  • Own a workload's end-to-end performance, find bottlenecks across application, data, scheduler, and GPU layers, and fix them in the relevant repository, leaving a benchmark behind.
  • Specify, generate or write, verify, and profile GPU kernels needed for physics campaigns and learned models.
  • Develop deep expertise in at least one area: multi-node AI training, inference serving at scale, or HPC simulation at scale.
  • Measure cluster performance on real jobs so queue policies and capacity decisions are based on observed results.

Benefits

  • Competitive compensation including salary and benefits.
  • Meaningful early-stage equity.
© 2026 Teal Labs, Inc
Privacy PolicyTerms of Service