Software Engineer, Systems ML

Meta•Menlo Park, CA

2d•$154,003 - $217,000

About The Position

Meta is seeking a Software Engineer to join the MTIA (Meta Training & Inference Accelerator) Software Tooling team, which develops and maintains the tooling ecosystem for Meta's in-house AI accelerator ASICs. The team provides debugging, profiling, memory analysis, and monitoring capabilities across the MTIA ecosystem, using Meta's full-stack ownership from silicon specifications to fleet observability. In this role, you will design and build developer tools across the MTIA tooling ecosystem, with a primary focus on debugging, sanitizer technology, and fault isolation for AI workloads running on MTIA hardware at scale. You will work across compilers, runtime, hardware, and ML frameworks, collaborating with cross-functional partners to deliver a high-quality developer experience for Meta's custom AI accelerators.

Requirements

Bachelor's degree in Computer Science, Computer Engineering, a relevant technical field, or equivalent practical experience.
4+ years of experience in systems software engineering, performance engineering, developer tooling, or a closely related field.
Experience building debugging, sanitizer, profiling, or diagnostic tools for complex software/hardware systems.
Proficiency in C++ and Python, including low-level systems programming and scripting for tool automation.
Experience working across multiple layers of a system stack, such as compiler, runtime, OS/driver, and hardware.
Track record of leading the technical design and delivery of tooling or infrastructure projects through production deployment.
Experience debugging issues across application, runtime, and hardware boundaries and tracing issues across those boundaries.

Nice To Haves

Demonstrated ongoing AI skill development, such as prompt/context engineering or agent orchestration, and staying current with emerging AI technologies.
Experience implementing responsible and ethical AI practices, such as risk assessment, bias mitigation, and quality and accuracy reviews.
Experience with accelerator ecosystems such as GPU/CUDA, TPU, or custom ASICs, including memory analysis, performance profiling, and runtime debugging with toolchains such as cuda-gdb, nsight-compute, and nsight-systems.
Familiarity with ML framework internals, including PyTorch graph execution, torch.compile, and operator dispatch, and AI compiler stacks such as MLIR, LLVM, TVM, and Triton.
Demonstrated ability to integrate AI tools to optimize or redesign workflows and drive measurable impact, such as efficiency gains or quality improvements.
Experience with Linux kernel and driver-level debugging, hands-on sanitizer and dynamic-analysis technology, or building comparable memory-checking tools.
Experience at the device-software boundary, including accelerator runtime/driver internals, DMA and memory-mapped device behavior, on-device fault handling, or firmware-assisted error reporting.
Experience with Linux debugging and profiling infrastructure, including gdb, perf, eBPF, ftrace, core dump analysis, and hardware performance counters; familiarity with binary formats and debugging metadata such as ELF/DWARF.
Experience with compiler-based instrumentation and runtime shadow-memory techniques underlying sanitizers, such as LLVM instrumentation passes, interceptors, and allocator red zones.
Track record of building developer tools adopted by large engineering populations, ideally with contributions to open-source projects such as LLVM sanitizers, gdb, Valgrind, or Triton.
Bachelor's degree plus 5+ years, Master's degree plus 3+ years, or PhD plus 2+ years in systems software, developer tooling, or accelerator software development.

Responsibilities

Lead the design and development of MTIA profiling and debugging tools, including performance profiling, live debugging, and core dumps for accelerator workloads.
Build tooling for performance profiling, kernel-level diagnostics, and multi-rank fault isolation.
Contribute to MTIA software tooling infrastructure for profiling, performance debugging, memory analysis, monitoring, and reliability analysis for training and inference workloads.
Own significant components end-to-end, from technical design through implementation, production rollout, and long-term maintenance.
Collaborate with MTIA compiler, runtime, kernel, and hardware teams to instrument the software stack with required hooks, sanitizers, and debuggers.
Drive AI-native tooling approaches, using automation and LLM-guided diagnostics to reduce time to root cause.
Partner with internal product teams across advertising, recommendations, and generative AI to understand developer pain points and prioritize tooling investments.
Communicate architectural decisions through design documents and cross-team reviews, and share debugging methodology and best practices with engineers across the MTIA stack.