What You'll Do

Design eval suites with AI agents in the loop. For every model release, curate a thoughtful mix of advanced, basic, long-context, and customer-use-case-specific evals. Use Claude to generate, validate, and prune candidate test cases at speed.
Build custom evals for target customers by orchestrating AI agents to mine trajectories from their workloads and synthesize representative eval sets.
Automate eval execution end-to-end with AI-driven pipelines on top of standard tooling (Docker, Git, CI). The goal is a system that runs itself between releases, not a script you re-run by hand.
Build automations to forecast and benchmark model performance on Cerebras for our top customers, including modeling how fast customer-specific workloads will run in production.
Build product-quality tooling that synthesizes quality + performance data into a single, easy-to-use view.

Skills & Qualifications

Experience building AI agents. You ship real systems with Claude (or equivalent) as a force multiplier. You've built things that would have been infeasible solo without AI agents in the loop.
Strong math/stats background.
Comfort with Docker, Git, and the standard automation stack.
A taste for tooling design. You've shipped something that a non-engineer used without complaining. Bonus if AI helped you ship it.

Performance-tuning experience on custom silicon, GPUs, or FPGAs.
Experience designing evals for agentic / coding / long-context / multimodal use cases.
Familiarity with open-source eval frameworks (EvalScope, lm-eval-harness, etc.).