C3 AI Data Scientist Interview Experience — The Coding Round Did Not Match Earlier Reports

Process:

OA → HR BQ → ML Case → ML Knowledge → Coding, for four interview rounds in total. I should be done here.

OA:

The same questions as the 2023 version on the forum. If you read through the interview reports here from beginning to end, you should basically be fine.

HM BQ:

The questions went into more detail than I expected. There were a lot of self-evaluation questions, such as what I thought my strengths and weaknesses were, how my manager would evaluate me, how my colleagues would evaluate me, and what I learned from a particular experience. I suggest preparing several concrete examples in advance because the interviewer will follow up on the details.

ML Case:

The problem was waste caused by a supermarket overstocking fresh food. It was open-ended: I walked through how I would collect the data, clean it, choose a model, and evaluate it. The interviewer interacted throughout and inserted questions while I was talking. The retail-inventory case previously shared on the forum follows the same pattern.

ML Knowledge:

This was rather different from what people had written on the forum. It was not a rapid-fire test of memorized knowledge. Instead, I was asked to explain one end-to-end ML project from my resume: how I did it from beginning to end, what the difficult parts were, how I solved them, and how I explained the model's results to nontechnical people. It felt more like a deep dive into my resume than a knowledge test. It ended after ten minutes.

Coding, completely different from the existing interview reports:

They gave me LeetCode 160 and one graph-theory problem. It was not the kind of precision-and-recall question, Two Sum variation, hand-built cross-validation, SGD, or plotting task described in the forum reports. I prepared for MLOps, data processing, and visualization based on those reports, and what I got did not match at all.

My advice for the people coming after me: do not prepare for the coding round using only the existing forum reports. You should still practice ML-related data processing, but you cannot skip the LeetCode basics either: linked lists, binary trees, graph BFS and DFS, and common medium problems. You need to cover both sides. The signals from the first three rounds and the job description do not mean the coding round will be ML-focused.

Has anyone interviewed for the same role recently? I want to know whether the coding-round question set really varies this much from one interviewer to another.