About the RoleYou will build the model runtime within the inference engine that executes complex, frontier models at scale on OpenAI's custom silicon. The runtime will sit between models running on the hardware and the upper layers of the cluster serving software stack, translating demanding inference workloads into efficient execution while optimizing for throughput, latency, utilization, and reliability.
You will work across model architecture, distributed systems, compilers, kernels, and silicon to design a production-grade runtime comparable in ambition to systems such as vLLM and SGLang, but customized and optimized for OpenAI's AI accelerator. Your work will shape how new model capabilities map onto the platform and how quickly custom silicon can deliver meaningful performance in production.
In this role, you will:- Design and implement the LLM inference runtime for frontier models running on custom silicon.
- Build scheduling, continuous batching, memory management, KV-cache management, and execution orchestration for high-performance inference.
- Develop distributed execution strategies across chips, hosts, and racks, including model partitioning, communication, and synchronization.
- Optimize end-to-end latency, throughput, memory efficiency, and hardware utilization across diverse model architectures and serving workloads.
- Partner with kernel, compiler, architecture, and silicon teams to co-design interfaces and remove performance bottlenecks across the stack.
- Enable new model features, execution patterns, numerical formats, and hardware capabilities in a reliable production runtime.
- Create profiling, observability, benchmarking, and performance-modeling tools that make runtime behavior measurable and actionable.
- Debug complex correctness, performance, and reliability issues spanning model code, runtime software, communication layers, and hardware.
- Turn workload insights into clear requirements for future generations of silicon and system architecture.
You might thrive in this role if:- Have strong systems programming experience in C++, Rust, Python, or comparable performance-oriented environments.
- Have built or optimized runtimes, distributed systems, compilers, kernels, model-serving infrastructure, or adjacent systems software.
- Understand modern LLM inference, including prefill and decode behavior, batching, KV-cache tradeoffs, and model parallelism.
- Can reason quantitatively about latency, throughput, compute intensity, memory bandwidth, communication, and utilization.
- Are comfortable profiling and debugging performance across multiple layers of a hardware-software stack.
- Can design clean abstractions while retaining the low-level control needed to extract performance from specialized hardware.
- Work effectively across model, systems, compiler, kernel, and hardware teams to drive ambiguous technical problems to closure.
- Care about production quality, including correctness, observability, reliability, maintainability, and graceful behavior at scale.
To comply with U.S. export control laws and regulations, candidates for this role may need to meet certain legal status requirements as provided in those laws and regulations.
