How do you combine code tests with evaluating model output?
Answer out loud or in your head first, then compare.
Answer
I check code with ordinary tests: does the parser reject bad data, does the limit work, does a retry avoid creating a second record. I assess the quality of a model's answers or classifications on prepared examples with set expectations. These are different checks.
If the fake model response in a test is always perfect, I'll check how such a response is handled, but not the model's real quality. I also need cases of refusal, error and incomplete output. After finding a real bug, it's worth adding an example that keeps it from coming back. You don't need to run an expensive model in every small code test.
Why it matters: Shows what independent verification of AI changes means.
Possible follow-ups: Which tests run on every change, and which separately?
How did it go? Be honest — “Not yet” puts the question on your review list.
Start with
- B-P01 · What is Node.js, and how is it different from JavaScript in the browser? — JavaScript, TypeScript and Node.js
- B-P04 · What is a Promise? — JavaScript, TypeScript and Node.js
- B-P10 · What does TypeScript give you, and what doesn't it? — JavaScript, TypeScript and Node.js
Further reading
- RAS Tech: Our Recruitment Process — All You Need to Know(opens in a new tab)Process; accessed 22.10.2026. A general description of the technical stage, internet access and AI; no guarantee of your round's format.