About the role

The Cloud Inference team scales and optimizes Claude to serve the massive audiences of developers and enterprise companies across AWS, GCP, Azure, and future cloud service providers (CSPs). We own the end-to-end product of Claude on each cloud platform, from API integration and intelligent request routing to inference execution, capacity management, and day-to-day operations.

Within Cloud Inference, the model & inference launch team owns the validation pipeline for our inference server and load balancer on these platforms. We're responsible for every inference change — model launches, performance improvements, safeguard integrations — landing on cloud platforms with correctness, performance, and reliability intact.

This is high-leverage infrastructure work: validation has to be fast and cheap enough to run on the same accelerators that serve customers, trustworthy enough to replace manual checks, and consistent enough that a change working on Anthropic first-party means it works everywhere. This directly determines how fast frontier models and features ship to every cloud platform, and how quickly performance wins reach production — reclaiming capacity at a time when compute is our scarcest resource.

Key responsibilities

• Be on the critical path for frontier model launches, bringing up inference for new model architectures and shipping them to cloud platforms in lockstep with our first-party platform
• Work with the core inference team to bring new inference features (e.g. structured sampling, prompt caching, and more) to cloud platforms, owning the platform-specific integration that gets them to production
• Identify and dive deep on the gaps that make inference behave differently across first-party and CSPs — config drift, observability, deployment patterns, hard cross-platform bugs — and fix them at the source rather than building platform-specific workarounds
• Design, build, and own the CI/CD infrastructure for the inference server and load balancer across cloud platforms, with shadow traffic, performance baselines (throughput and latency), and correctness checks that catch regressions before production
• Drive down merge-to-production cycle time by making validation faster, more parallel, and cost-effective enough to run on the same constrained accelerator pool that serves customers, without trading away reliability 
• Analyze observability data across providers to identify performance bottlenecks, cost anomalies, and regressions, and drive remediation based on real-world production workloads

Minimum qualifications

• Have a strong interest in LLM serving; prior inference or ML experience is not required 
• Have significant software engineering experience, with a strong background in high-performance, large-scale distributed systems serving millions of users
• Have a track record of building automation or test infrastructure that measurably improved release velocity or reliability
• Have experience building or operating services on at least one major cloud platform (AWS, GCP, or Azure), with exposure to Kubernetes, Infrastructure as Code, or container orchestration
• Thrive in cross-functional collaboration with both internal teams and external partners
• Are a fast learner who can quickly ramp up on new technologies, hardware platforms, and provider ecosystems
• Are highly autonomous and take ownership of problems end-to-end, including work that falls outside your job description

Preferred qualifications

• LLM inference optimization, batching, and caching strategies
• Capacity-constrained scheduling or shared-resource test infrastructure
• Solid understanding of multi-region deployments, request routing, load balancing, global traffic management
• Working with CSP partner teams to scale infrastructure across multiple platforms, navigating differences in networking, security, privacy, and managed service
• Proficiency in Python or Rust