Software Engineer - Evals

xAI xAI · AI Frontier · Palo Alto, CA · Engineering

Software Engineer for the Evals team at xAI, responsible for building and maintaining systems to measure AI model capabilities and behaviors, including dataset design, grading schemes, and execution infrastructure. The role focuses on creating tools that accelerate frontier research and diagnose agent failures.

What you'd actually do

  1. The evals team creates and executes instruments for measuring model capabilities and behaviors.
  2. These instruments are used to clean datasets, provide feedback for experiments, and track growth throughout training.
  3. The evals team also diagnoses why agents fail and how to make improvements both short and long term.
  4. We manage the execution infrastructure, design data sets and grading schemes.
  5. You will design, build, and ship delightful and bulletproof systems that engineers love to use every day.

Skills

Required

  • Software engineering proficiency
  • 3+ years of experience actively writing code
  • A track record of shipping high quality projects
  • strong sense of urgency
  • Exceptional product taste
  • Strong attention to detail

Nice to have

  • Previously founded a startup or joined a fast-growing early-stage startup
  • Familiarity with AI/ML workflows, annotation tools, or evals
  • Experience creating and maintaining component libraries, design systems, or internal tools

What the JD emphasized

  • mission-critical research platform
  • measuring model capabilities and behaviors
  • diagnoses why agents fail
  • design data sets and grading schemes
  • bulletproof systems that engineers love to use
  • Exceptional product taste
  • translate fuzzy notions of quality, helpfulness, truthfulness, and safety into concrete, measurable criteria

Other signals

  • evals team creates and executes instruments for measuring model capabilities and behaviors
  • diagnoses why agents fail and how to make improvements
  • manage the execution infrastructure, design data sets and grading schemes
  • design, build, and ship delightful and bulletproof systems that engineers love to use