OpenAISan Francisco, CAEvaluates new hardware platforms by porting benchmarks and workloads, analyzes performance across compute/memory/networking, identifies bottlenecks, and optimizes for AI systems. Requires expertise in performance analysis, system architecture, and debugging across hardware/software boundaries.
Defines rack- and cluster-level reference architectures for AI infrastructure, translates workload requirements into designs, collaborates with partners and modeling teams to evaluate tradeoffs, and drives vendor roadmaps to address technology gaps.
Ensure reliability of large GPU supercomputing clusters by diagnosing hardware/firmware/OS issues, automating monitoring, driving firmware rollouts, and working directly with vendors.
Own and debug multi-thousand-GPU network fabric (RDMA/RoCE, NVLink/NVSwitch) for large-scale AI training and inference. Requires backend language proficiency, large-scale cluster experience, and cross-stack ownership.
Performance engineer focused on cross-layer investigations of Anthropic's inference fleet for Claude, optimizing throughput, latency, reliability, and correctness while building observability and partnering with kernel and serving teams.
Site Reliability Engineer drives end-to-end reliability for AI fine-tuning platform Tinker, including SLOs, monitoring, incident response, and multi-tenant GPU scheduling. Requires distributed systems experience, software proficiency for reliability, and production incident handling.