Scale AI Interview Process - New Grad Software Engineer

I had applied for Scale AI (without referral), got the OA link. I had two rounds after the OA:

1. Debugging Practical (Python)

As mentioned in multiple places, the codebase involves projects, tasks, and courses. It's a small repo with multiple files, and you need to navigate through them, identify bugs in different functions, and fix them.

There were about 7 to 8 bugs in total, split into 3 parts with 2–3 bugs each. You essentially need to get the 3 test cases passing.

A few things I remember:

• Checking conditions like headcount being greater than 0.

• Sorting order, specifically sorting based on a class variable.

• Checking whether you're using the correct ID (course ID vs. ID).

• The third part involved deep copy vs. shallow copy, so definitely brush up on that.

I was able to pass all three test cases. Overall, I found the debugging round manageable.

2. Backend Practical (Python, AI-assisted)

This round had two questions. I don't remember the exact problem statements, but they involved things like LLM calls, session orchestration, authorisation, and other backend concepts.

There was quite a lot to implement, and honestly, I found it difficult to see how someone could fully code both questions in 45 minutes.

Now, regarding AI usage, since a lot of people have asked me about this:

Initially, the interviewer mentioned that AI wasn't allowed. I clarified that the recruiter had explicitly told me AI tools were permitted, and she was then okay with it.

I used Claude. My approach was to explain my entire system design, logic, and how I wanted to solve the problem, and then use AI to generate the implementation. I reviewed the code, discussed time and space complexity, edge cases, and explained my decisions throughout.

I personally felt the round went reasonably well, but I wasn't moved forward to the HM + Credo rounds.

My personal guess (and this is purely speculation) is that the interviewer expected me to write most of the code myself and use AI primarily for syntax assistance or smaller implementation details, rather than generating the full implementation. I may have relied on AI more than she expected, even though I was driving the design and logic.

So my biggest suggestion would be to clarify the expectations around AI usage with your interviewer before starting, especially how much assistance is considered acceptable.

That's pretty much my experience. Hope this helps anyone preparing for the onsite, and good luck to everyone interviewing!