Evals Infrastructure Tech Lead / Manager

About the Role

Anthropic is at the forefront of AI research, dedicated to developing safe, ethical, and powerful artificial intelligence. Our mission is to ensure that transformative AI systems are aligned with human interests. We're looking for an experienced tech lead to join our Evals Infrastructure team, building the systems that let us measure what our models can actually do. Evaluation is how we know whether a model is safe to ship — you'd own the infrastructure that makes those measurements fast, reliable, and trustworthy at scale. In this role you'll work at the intersection of inference, research and infrastructure engineering: managing the large scale distributed systems that orchestrate evals for our frontier models, building and scaling the harnesses researchers use to design and run evals, making results reproducible and interpretable, and ensuring eval signal is available where decisions get made. Your work directly shapes what we build and what we don't.

• Lead the team building the distributed systems that schedule, orchestrate, and execute evals for our frontier model training
• Own eval throughput and cost: compute allocation across suites, queueing against constrained accelerator pools, caching and reuse of eval work
• Build and scale the harnesses researchers use to define, run, and iterate on evals
• Make eval results trustworthy — determinism, reproducibility, and honest uncertainty quantification on reported metrics
• Ensure eval signal reaches the dashboards and reviews where launch decisions actually get made
• Contribute directly as an engineer while managing and growing the team, prioritizing its work, and coaching your reports

• Have led technical projects end-to-end on large-scale distributed systems, and have 1+ years managing engineers (or tech-lead-with-reports experience)
• Are strong in Python and Rust
• Have built high-throughput, fault-tolerant systems on cloud or on-prem accelerator fleets
• Care about measurement quality, not just pipeline uptime — you'd notice if a metric moved for the wrong reason
• Communicate well with researchers and can translate research needs into infrastructure
• Are deeply interested in the transformative effects of advanced AI and committed to safe development

• Worked on LLM inference or training infrastructure
• Experience with eval or benchmarking systems, especially agentic evals requiring sandboxed execution
• Working statistical literacy — variance, confidence intervals, sample-size sufficiency for noisy metrics
• Experience with observability and regression detection over time-series metrics

• Rebuilding the eval orchestration layer to cut wall-clock time on the pre-train eval suite
• Designing compute allocation and scheduling so eval suites fit inside a fixed fraction of a production run's chip-hours
• Adding rigorous uncertainty estimates to top-line dashboard metrics so checkpoint-to-checkpoint comparisons are actually decision-grade
• Building sandboxed execution infrastructure for agentic evals