Performance Engineer, Inference Systems

About this role

Anthropic seeks a Performance Engineer to optimize and validate their large-scale Claude inference system across throughput, latency, reliability, and correctness. You'll investigate performance gaps across the entire stack, own correctness evaluation pipelines, and partner with infrastructure teams to drive high-impact optimizations.

What you'll do

• Conduct cross-layer performance investigations to identify gaps between actual fleet performance and theoretical limits
• Own and improve the correctness evaluation pipeline that validates model outputs across hardware platforms and configurations
• Build observability tools, dashboards, and models to make performance metrics and their interactions visible across the stack
• Partner with kernel, serving, and infrastructure teams to prioritize and implement optimization opportunities
• Investigate performance regressions and root causes in production inference systems
• Stack-rank optimization opportunities by impact and effort to guide team priorities

What they're looking for

• Performance engineering and profiling in production systems
• Python and ability to work with large codebases
• Data analysis tools such as SQL and pandas
• Roofline analysis and latency/throughput optimization
• ML inference systems and LLM serving infrastructure
• GPU/TPU accelerator performance concepts
• Distributed systems reliability engineering
• Technical communication and influence through evidence