Senior / Staff AI Inference Infrastructure Software Engineer

We're looking for a senior engineer to own and advance the core of our AI inference infrastructure — the systems that serve large models at scale. (We focus exclusively on inference; we don't do training.) You'll set technical direction in one or more of our focus areas (inference serving, ML compilers, GPU kernels, communication libraries), ship high-impact performance work, and raise the bar for the whole team. This is a deep technical role with ownership and influence. We expect you to go from a research idea down to PTX, and to make architectural decisions that move cluster-level inference cost and performance. What You'll Do Own a critical part of the stack end-to-end and drive its roadmap: Inference / serving: continuous batching, paged/prefix KV cache, speculative decoding, prefill/decode disaggregation, quantization (FP8/INT8/INT4). Compilers: MLIR / TVM / XLA / TorchInductor passes, fusion, layout selection, autotuning, codegen for GPUs/accelerators. Kernels: CUDA / CUTLASS / Triton, Tensor Cores (WMMA/WGMMA), FlashAttention- class kernels, Hopper/Blackwell features (TMA, warp specialization). Communication: NCCL internals, custom all-reduce/all-to-all, RDMA/InfiniBand, NVLink/NVSwitch topology, comm/compute overlap, tensor / pipeline / expert parallelism for inference. Deliver measurable wins in latency, throughput, cost, and reliability at scale. Set engineering standards: perf reviews, benchmarking, regression gates. Mentor engineers and provide technical leadership across teams. Represent the team in design reviews and (optionally) externally via talks/writing.

7+ years building high-performance systems software (or equivalent depth). Expert-level C++ and CUDA ; strong systems fundamentals. Deep expertise in at least one focus area (serving, compilers, kernels, or communication) and working knowledge of the others. Track record of shipping performance-critical systems to production and owning them operationally. Experience profiling and optimizing GPU workloads with measurable, significant impact. Preferred Qualifications Experience with high-QPS, low-latency LLM inference serving at scale. Contributions to major ML systems / open-source projects (vLLM, TensorRT-LLM, SGLang, Triton, CUTLASS, NCCL, PyTorch, MLIR/TVM/XLA). Experience with low-precision (FP8/INT4) and modern GPU architectures (Hopper/Blackwell). Prior technical leadership / mentoring of engineers. M.S. / Ph.D. in CS/CE or related, or equivalent industry depth. Nice to Have Publications or patents in ML systems / HPC. Conference talks or technical writing. Experience bringing up new hardware or custom accelerators. What We Offer Ownership of frontier-scale AI infrastructure with direct production impact. A strong, senior-heavy team and real technical autonomy. Highly competitive compensation, significant equity, and benefits.