Senior Compiler Engineer (Triton/Mojo)

General Diffusion’s mission is to decouple intelligence from silicon. We believe the path to AGI requires a universal translation layer that makes compute fungible across any architecture—from H100s to TPUs to neuromorphic chips.

About the role

As a Senior Compiler Engineer, you will be the architect of CG1 (Code Generator Agent), the engine responsible for emitting highly optimized kernels for diverse hardware backends. You will work at the intersection of ML compilers (MLIR, Triton) and low-level systems programming (Mojo, C++, CUDA) to build a JIT engine that rivals hand-tuned performance.

What you might work on

• Architecting the CG1 backend to translate our General Intermediate Representation (GIR) into device-specific ISA.
• Implementing custom Triton passes to optimize memory access patterns for next-gen HBM3e architectures.
• Building the first production-grade Mojo transpiler for CPU fallback and heterogeneous orchestration.
• Designing automated correctness fuzzers to validate generated kernels across NVIDIA, AMD, and Google silicon.
• Collaborating with hardware partners to support pre-release silicon (B200, MI300, Trillium).

What we’re looking for

• Deep expertise in compiler infrastructure (LLVM, MLIR, TVM, or XLA).
• Production experience with OpenAI Triton or Modular Mojo.
• Strong understanding of GPU architecture (memory hierarchy, warp scheduling, tensor cores).
• Ability to write and debug raw CUDA/PTX when necessary.
• A "physics-first" approach to optimization—you care about cycles, joules, and nanoseconds.

Our culture

• Silicon Neutrality. We build for the world where compute is a commodity, not a monopoly.
• Radical Efficiency. We believe software bloat is an existential risk to AGI.
• Deep Work. We value long periods of uninterrupted focus over endless meetings.