Our agents do real accounting work on live client books, so "it looks right" is not good enough. As an AI Test Engineer, you will build the evals that measure whether they get it right: the test cases, the scoring and the systems that run them on every change.
You will work closely with our research team, with our Applied Accounting Researcher, who decides what correct looks like, and with the engineers building the AI CFO and our agents. You will be based in our Copenhagen office.
Build and grow our eval suites across accounting tasks, from reconciliation to reporting and consolidation
Turn real cases from accounting firms into test cases with clear, agreed answers
Design scoring you can trust, using model-graded checks only where they hold up
Run evals on every model, prompt and tool change, and catch regressions before release
Track agent quality over time, and make the results easy for the whole team to read
Help decide when an agent is good enough to take on a new task
5 or more years of experience in software engineering, test engineering or machine learning
You have built a test harness for LLMs or agents before: the datasets, the runners and the scoring
A strong software engineer: your test systems are production code, not a folder of scripts
Rigour: you design tests that are hard to fool
Clear writing about what results mean, and how sure you are of them
Statistics, and experience measuring noisy systems
Experience building CI pipelines
An interest in accounting or finance
This job comes with several perks and benefits
