Principal AI Accelerator Tools Development Engineer
Core
Design and develop scalable stress, performance, and validation frameworks for MAIA AI accelerator platforms, building workload generation infrastructure to exercise compute, memory, interconnect, and system-level resources.
Role type
Principal AI Accelerator Tools Development Engineer
Builds
Scalable stress, performance, and validation frameworks; workload generation infrastructure; automated infrastructure for deployment, orchestration, and telemetry; benchmarking methodologies and performance dashboards.
Domain
AI accelerator hardware (MAIA), high-performance computing (HPC), distributed AI systems, compiler technologies.
Deliverable
production ML models | infrastructure
Required skills
C++, PyTorch, Triton, Python, custom MAIA SDKs, kernel development, performance profiling, bottleneck analysis, distributed AI infrastructure, compiler toolchains (LLVM, MLIR), silicon bring-up, platform qualification, regression testing, thermal/power analysis.
Preferred skills
Experience with AI compiler technologies and kernel generation frameworks, training/optimizing large-scale AI models (LLMs), custom AI accelerator SDKs, collective communication libraries, cloud-scale validation infrastructure.
Technologies
PyTorch, Triton, Python, C++, LLVM, MLIR, MAIA SDKs.
Responsibilities
Design and develop scalable stress, performance, and validation frameworks for MAIA AI accelerator platforms; Build workload generation infrastructure capable of exercising compute, memory, interconnect, networking, storage, and system-level resources; Develop and optimize kernels targeting custom AI accelerators; Analyze execution behavior across the hardware-software stack and identify bottlenecks impacting utilization and performance; Collaborate with compiler and runtime teams to improve workload efficiency and hardware utilization; Build automation around model compilation, kernel validation, regression testing, and workload portability.
Seniority
Principal, hands-on IC with strategy & mentorship
