Senior AI Compiler Engineer for GPUs & Accelerators

Oho Group; United States; On-site; USD 180,000 - 280,000; Full time

Job summary

Oho Group is an early-stage AI systems company hiring a Principal Compiler Engineer to help bring up and optimise AI models across new hardware targets.

You will develop compiler backends for GPUs, NPUs and custom AI accelerators, build MLIR and LLVM lowering pipelines, and bring up training and inference workloads from kernels upwards. You will analyse code and hardware to identify bottlenecks and optimise Triton kernels while integrating with PyTorch and accelerator runtimes.

Qualifications

• Strong professional experience with LLVM, MLIR or comparable compiler infrastructure.
• Excellent C++ and strong systems-level debugging ability.
• A strong understanding of performance across compute, memory and parallel execution.
• The ability to own complex technical problems within a small, fast-moving company.

Responsibilities

• Develop compiler backends for GPUs, NPUs and custom AI accelerators.
• Build MLIR and LLVM lowering pipelines, transformation passes and target-specific code generation.
• Bring up modern training and inference workloads from the kernels upwards.
• Analyse generated code and hardware behaviour to identify performance bottlenecks.
• Optimise Triton kernels across compute, memory movement, data layout, scheduling and parallel execution.
• Integrate compiler and kernel infrastructure with PyTorch, vLLM and accelerator runtimes.

Skills

Performance tuning

Systems debugging

Parallel execution

Problem ownership

Tools

LLVM

MLIR

C++

Job description

Oho Group is an early-stage AI systems company hiring a Principal Compiler Engineer to help bring up and optimise AI models across new hardware targets.

You will develop compiler backends for GPUs, NPUs and custom AI accelerators, build MLIR and LLVM lowering pipelines, and bring up training and inference workloads from kernels upwards. You will analyse code and hardware to identify bottlenecks and optimise Triton kernels while integrating with PyTorch and accelerator runtimes.