Evaluation Engineer, Applied AI

The Role

This is a chance to build the frontier LLM evals that shape the next generations of flagship models.

Could a model design a computer chip, or find the backdoor an attacker left in a network? Public benchmarks can't tell us, as they're generally not a good measure of real work.

As an Evaluation Engineer, you'll own evals from start to finish. You'll run workshops with customers, industry experts and our customer-facing teams to understand their most important industry use cases (e.g., an engineer updating a part in CAD software), and help decide which are worth building first. You'll also look beyond today's use cases and design moonshot evals for work no model can do yet, so Mistral gets there first. Then you'll turn each one into a benchmark, often one where an agent has to do the work inside real software. You'll source the data, decide how to score it, and run it on every new Mistral model and on competitor models.

When our models fall short, you'll work with the post-training teams to close the gap. When they catch up, you'll make the eval harder. You'll write up and present your results so senior leaders can act on them. They also give our go-to-market teams evidence of where Mistral leads.

You'll learn new fields quickly. One month you might be learning how chip designers check their work, and the next how a bank investigates a flagged payment.

The team is new and fast-paced, and you'll have a lot of autonomy. You'll spend a lot of time speaking with people across Mistral to work out what matters most, then decide what to build.

What You Will Do

• Run discovery with customers, industry experts and teams across Mistral, then help decide which use cases are worth building first.

• Design new benchmarks for industry use cases and frontier capabilities, including how they're scored.

• Build evals where agents work inside real software (e.g., CAD and simulation tools), and run them on our models with checks that the results stay reliable over time.

• Stress-test existing evals to find flaws and bugs before we rely on them.

• Source data with expert vendors. Where customer data can't leave the customer, build realistic proxy or synthetic data instead.

• Write up and present what you find so leaders across Mistral can act on it.

• Help other teams build their own evals, with tools and guides they can reuse.

Experience

• 2+ years of industry experience (e.g., in software, AI, or consulting), or a PhD.

• A software engineering or research background, with experience writing code that others can build on.

• Experience working directly with customers, industry experts, or other teams (e.g., through consulting, solutions engineering, or research collaborations).

• Experience building something with LLMs.

• Comfort working with large datasets, including building and cleaning them.

NICE TO HAVE

• Experience building AI evaluations, especially for industry applications.

• Published research on evals or contributions to open-source evaluation frameworks.

• Experience building datasets, generating synthetic data, developing labelling guidelines, or working with annotation vendors.

• A background in an industry outside software (e.g., engineering, finance, or science).

What We Offer

We offer a comprehensive benefits package designed to support your well-being, growth, and work-life balance. Benefits vary by country and may include healthcare coverage, parental leave, retirement plans, relocation support, wellness programs, meal and transportation allowances, and other location-specific perks.

For the most up-to-date details on benefits available in your location, please refer to our Benefits page.