Remote Compiler Engineer: AI Inference & GPU Optimizations

Hanzo; EE. UU.; A distancia; USD 150.000 - 210.000; Jornada completa

Ventajas ofrecidas por este puesto de trabajo

Remote-first

High autonomy

Opportunities to publish

Descripción de la vacante

Hanzo is seeking an experienced compiler engineer to develop technology that speeds AI model inference across cloud and edge environments. You will optimize computation graphs, lower neural network representations, and build hardware-specific execution paths for GPUs, TPUs, and accelerators.

Responsibilities include creating optimizing compiler components and graph transformations, generating efficient code and kernel calls, and improving memory management with techniques like KV caching and

Formación

• Strong C++ and/or Rust skills with deep compiler architecture knowledge.
• Familiarity with MLIR/LLVM and ML graph compilation tools (TVM, XLA, TensorRT, TorchDynamo).
• Understanding of GPU architecture, tensor cores, and warp scheduling.

Responsabilidades

• Create optimizing compiler components and graph transformation passes for large language model inference.
• Use compiler IRs to generate efficient machine code and kernel calls.
• Improve memory management and execution techniques, including caching and attention optimizations.
• Benchmark and profile performance with focus on throughput and cost per generated token.

Conocimientos

C++

Rust

Compiler architecture

GPU architecture

Performance measurement

Herramientas

MLIR

LLVM

TVM

XLA

TensorRT

TorchDynamo

Descripción del empleo

Hanzo is seeking an experienced compiler engineer to develop technology that speeds AI model inference across cloud and edge environments. You will optimize computation graphs, lower neural network representations, and build hardware-specific execution paths for GPUs, TPUs, and accelerators.

Responsibilities include creating optimizing compiler components and graph transformations, generating efficient code and kernel calls, and improving memory management with techniques like KV caching and