Software Engineer, TPU Compilers

Highlights

• Lead optimization of OpenAI's frontier models on TPUs across full forward pass.
• Improve latency, throughput and hardware utilization for state-of-the-art inference.
• Build benchmarks and tools to guide optimization and support new model architectures.

What they offer

• Equity

Ideal candidate

A systems-minded engineer who loves squeezing performance from accelerators and owns open-ended problems end-to-end. You enjoy profiling, collaborating with researchers, and turning low-level insights into production gains.

Requirements

• Practical experience optimizing ML workloads on TPUs via training, inference, or both
• Understanding of modern model architectures and mapping computation to accelerator hardware
• Ability to connect kernel-level behavior to full-model performance across the stack
• Strong software engineering skills with systematic profiling, debugging and validation
• Ownership of open-ended technical problems from prioritization to delivery
• Collaborative mindset for working with researchers and communicating tradeoffs
• Experience with quantization or attention kernels is especially useful