Performance Profiling Engineer: Build Chip-Level Profilers

Infinity Artificial Intelligence Institute; San Francisco (CA); On-site; USD 150,000 - 210,000; Full time

Job summary

Infinity is an early-stage AI infrastructure research company building software that makes non-NVIDIA chips competitive for AI inference. We’re hiring a performance-analysis engineer to build a profiler across chips, with per-operation attribution and automation that exposes bottlenecks.

Based in San Francisco, this full-time on-site role requires deep systems background, Python and Rust/C/C++, and experience building measurement tools when the docs are thin.

Qualifications

• Real performance-analysis experience profiling hard problems.
• Low-level systems background with counters, tracing, and instrumentation.
• Ability to build measurement tools when documentation is thin.
• Statistical care about noise and sample size before reporting results.
• Fluency in Python and at least one systems language (Rust, C, or C++).

Responsibilities

• Profiler-generation: build a profiler for a supported chip and ensure visibility across the stack.
• Per-operation attribution: break down an inference pass to show which operation costs time.
• Reimplement surface-area tooling per chip (nvprof/Nsight) for accelerators without profilers.
• Counter and telemetry discovery: identify signals and validate counters for performance.
• Faithful instrumentation: measure without perturbing timings.
• Bottleneck surfacing: translate measurements into exact blockers for optimization.

Skills

Performance-analysis

Low-level systems

Instrumentation

Python

Rust/C/C++

Tools

perf

VTune

Nsight

Job description

Infinity is an early-stage AI infrastructure research company building software that makes non-NVIDIA chips competitive for AI inference. We’re hiring a performance-analysis engineer to build a profiler across chips, with per-operation attribution and automation that exposes bottlenecks.

Based in San Francisco, this full-time on-site role requires deep systems background, Python and Rust/C/C++, and experience building measurement tools when the docs are thin.