Scale AI Software Engineer Interview Experience — Debug a Real Codebase, Then an LLM API Round

Phone Screen

It was the "party" question that's already been posted on the forum, pretty easy.

Debug

This round felt a bit harder. The question was the "assign project" one that's already on the forum, and the main issue was that time was tight. It's a new codebase with several files, and you need to go back and forth through it a lot (would've been a lot easier if I could've just used AI for this). Once you actually find the bugs, they're all easy to fix — things like "the for loop never calls the reset function" or "the filter should be on courseid instead of id", that kind of simple bug. The interviewer gave a fair amount of help along the way.

Practical

You need to call an LLM API once to solve the problem. It really comes down to how you use the OpenAI API and how you write the prompt. The follow-up was about doing some validation on the response, plus what you'd need to watch out for if you took this feature to production.

System Design

Also a question that had already been posted on the forum — similar to a system design question I'd seen in another thread here.

The last two rounds were both behavioral. Honestly I couldn't really tell what the difference between them was supposed to be. One of them asked the pretty common stuff — the project you're most proud of, feedback you've received, a time you had to work under a tight deadline. The other one asked about side projects, your strengths and weaknesses, that kind of thing.