Aurtiro
Senior Software Engineer – Compute Infrastructure (C++)
Aurtiro, San Francisco, California, United States, 94199
Senior Software Engineer – Compute Infrastructure (C++)
Build virtualization software that sits below CUDA and maximizes utilization across large GPU fleets. This is research-driven systems work: you'll design the low-level runtime, networking stack, and scheduling layer that turns raw accelerators into a fast, efficient compute platform for AI workloads.
The work involves deep performance optimization, novel isolation techniques, and squeezing every bit of efficiency from heterogeneous hardware. You'll touch the entire stack—from kernel‑level primitives to cluster orchestration—and work problems that go far beyond typical cloud infrastructure.
What you'll work on:
Virtualization layer
– Build the core software that isolates, schedules, and multiplexes GPU workloads across thousands of accelerators with minimal overhead.
Performance-critical systems
– Design high-performance C++ services for resource allocation, job dispatch, and data-path optimization. Profile end-to-end: memory layout, scheduling policies, network utilization.
Low-level networking
– Implement and optimize the stack below CUDA: RDMA, GPU-direct communication, and ultra-low-latency interconnects (NVLink, InfiniBand, RoCE).
Research & iteration
– Explore novel approaches to container startup, model loading, multi-tenancy, and failure handling. Many problems here don't have known solutions yet.
What we're looking for:
Strong systems fundamentals: concurrency, memory management, OS internals, networking.
Experience building performance-sensitive software in C++ (or Rust/Go if you're willing to work in C++).
Comfort with ambiguity—able to prototype, measure, debug hard problems, and iterate toward production-quality systems.
Excited by extremely complex debugging, profiling tools, and making things measurably faster.
Helpful but not required:
GPU computing (CUDA, ROCm, NCCL), accelerator runtimes, or HPC workloads.
High-speed networking or storage systems at scale.
Why this role:
Work at the boundary of what's possible: virtualization and scheduling problems with no established playbook.
Extremely broad scope—you'll touch every layer of a genuinely hard system, not just one service or API.
Small, exceptional team. The people here are among the best systems engineers you'll work with.
High leverage: the efficiency gains you unlock directly shape the economics of the platform.
Details:
We'll relocate for the right person. Open to engineers at all experience levels—if you're early career but have the raw ability and hunger for this kind of work, we want to hear from you.performance behavior.
Track record shipping and sustaining performance-focused infrastructure or runtime systems.
Fluency in diagnosing reliability + performance issues in distributed environments.
Preferred:
Scheduling/orchestration, capacity planning, resource accounting.
High-performance networking (RDMA/DPDK/kernel bypass) or equivalent systems depth.
Performance work involving accelerators or high-throughput I/O.
Logistics: Full-time,
in-office in downtown San Francisco . Competitive compensation and meaningful early-stage equity.
#J-18808-Ljbffr
The work involves deep performance optimization, novel isolation techniques, and squeezing every bit of efficiency from heterogeneous hardware. You'll touch the entire stack—from kernel‑level primitives to cluster orchestration—and work problems that go far beyond typical cloud infrastructure.
What you'll work on:
Virtualization layer
– Build the core software that isolates, schedules, and multiplexes GPU workloads across thousands of accelerators with minimal overhead.
Performance-critical systems
– Design high-performance C++ services for resource allocation, job dispatch, and data-path optimization. Profile end-to-end: memory layout, scheduling policies, network utilization.
Low-level networking
– Implement and optimize the stack below CUDA: RDMA, GPU-direct communication, and ultra-low-latency interconnects (NVLink, InfiniBand, RoCE).
Research & iteration
– Explore novel approaches to container startup, model loading, multi-tenancy, and failure handling. Many problems here don't have known solutions yet.
What we're looking for:
Strong systems fundamentals: concurrency, memory management, OS internals, networking.
Experience building performance-sensitive software in C++ (or Rust/Go if you're willing to work in C++).
Comfort with ambiguity—able to prototype, measure, debug hard problems, and iterate toward production-quality systems.
Excited by extremely complex debugging, profiling tools, and making things measurably faster.
Helpful but not required:
GPU computing (CUDA, ROCm, NCCL), accelerator runtimes, or HPC workloads.
High-speed networking or storage systems at scale.
Why this role:
Work at the boundary of what's possible: virtualization and scheduling problems with no established playbook.
Extremely broad scope—you'll touch every layer of a genuinely hard system, not just one service or API.
Small, exceptional team. The people here are among the best systems engineers you'll work with.
High leverage: the efficiency gains you unlock directly shape the economics of the platform.
Details:
We'll relocate for the right person. Open to engineers at all experience levels—if you're early career but have the raw ability and hunger for this kind of work, we want to hear from you.performance behavior.
Track record shipping and sustaining performance-focused infrastructure or runtime systems.
Fluency in diagnosing reliability + performance issues in distributed environments.
Preferred:
Scheduling/orchestration, capacity planning, resource accounting.
High-performance networking (RDMA/DPDK/kernel bypass) or equivalent systems depth.
Performance work involving accelerators or high-throughput I/O.
Logistics: Full-time,
in-office in downtown San Francisco . Competitive compensation and meaningful early-stage equity.
#J-18808-Ljbffr