Compiler Engineer

Role overview Build compiler and runtime capabilities that make neural-network inference faster and more efficient. The work spans graph optimization, intermediate-representation lowering, and hardware-specific code generation for GPUs, TPUs, and newer accelerator types.

Responsibilities - Develop compiler optimizations and graph transformation passes for language-model inference. - Use compiler intermediate representations to generate efficient machine code and kernel calls. - Improve memory handling and inference techniques such as key-value caching, speculative decoding, and paged attention. - Benchmark and profile performance, tracking throughput and cost.

Requirements - Strong C++ or Rust skills and a deep understanding of compiler architecture, including intermediate representations, optimization passes, and backends. - Familiarity with MLIR, LLVM, or machine-learning compilation tools such as TVM, XLA, TensorRT, or TorchDynamo. - Understanding of GPU architecture, including tensor cores, shared memory, and warp scheduling. - Careful approach to performance measurement, micro-benchmarks, and regression testing.

Nice to have - Experience writing CUDA or Metal kernels for attention or fused operations. - Contributions to open-source compiler or runtime projects.

Benefits and work setup - Remote-first work across distributed locations, with autonomy and opportunities to contribute publicly to open-source projects.