Member of Technical Staff, GPU Compiler

About the role

We are building the world’s fastest GPU compiler, breaking traditional correctness constraints to enable aggressive search‑based optimization. As a Member of Technical Staff for GPU Compiler Engineering, you will create the machine that performs the search and delivers high‑performance kernels across multiple hardware targets.

Key responsibilities

• Design and extend MLIR dialects and passes for training and inference workloads.
• Develop the LLVM backend below ptxas, handling instruction selection, scheduling, register allocation, and direct cubin emission for NVIDIA, AMD, TPU, and Trainium.
• Expand search‑based compiler infrastructure with agent‑ and RL‑driven program search and formal correctness proofs.
• Implement classic compiler optimizations tuned for large‑scale training and create hybrid codegen paths when direct MLIR lowering is impractical.
• Own testing, benchmarking, and performance regression systems, including bit‑identical hardware models.

Required profile

• Deep experience with compiler infrastructure such as LLVM or MLIR.
• Strong background in GPU architecture and low‑level optimization (CUDA, ROCm, PTX/SASS, GCN/RDNA).
• Hands‑on experience with at least one GPU ISA and familiarity with ML compiler stacks (XLA, TVM, Triton, torch.compiler).
• Solid systems programming skills in C++ and/or Rust.
• Proven track record of building production‑grade compiler infrastructure.

Required skills

• LLVM
• MLIR
• CUDA
• ROCm
• PTX / SASS
• GCN / RDNA assembly
• XLA
• TVM
• Triton
• torch.compiler
• C++
• Rust
• Reinforcement Learning (RL) for program search
• Formal verification / SMT solvers
• StableHLO / (Stable)HLO

What we offer

• Opportunity to work on the fastest GPU compiler with direct impact on AI compute costs.
• Small, frontier‑scale team that has pre‑trained foundation models on thousands of GPUs.
• Relocation assistance and an onsite office in San Francisco.
• Competitive base salary $285,000–$315,000 plus equity and benefits.