NAVER Place AI Internship: Edge Cases Decided the Test, the Job Description Decided the Interview
Loading comments...
I went through the NAVER Place AI development internship process from the coding test to the technical interview. I had passed the resume screen for two positions, but the rules allowed only one, so I chose Place AI.
The test questions stay private under the assessment rules. This is about what decided the score, where I got stuck in the interview, and what I will do first next time.
Coding test: the inputs to guard before the answer
Three problems on Codility: implement a machine-learning method without libraries, build a short pipeline from preprocessing to training, and an algorithm problem where time complexity mattered.
All three pass quickly on ordinary inputs. The difference came after that.
- Division by zero. Distances and ratios needed the zero-denominator input considered first.
- Hidden missing values and skewed labels. A clean-looking table still had gaps, and a lopsided label split inflated accuracy easily.
- Large inputs. Code that passed the small examples timed out on large ones and had to be rewritten as a single pass.
Knowing the textbook path was not enough; thinking of the strange input first was closer to "implemented correctly."
Technical interview: what I prepared was not what came
About 45 minutes on Zoom. I had prepared ML fundamentals, my projects, and LLM concepts. The weight was the other way around.
| Area | How much I prepared | Share of the interview |
|---|---|---|
| ML and LLM concepts | a lot | a few short questions |
| Project defense | prepared | one or two |
| Evaluation scenario design | almost none | over half |
Concept and project questions went fine, including a follow-up on how I measured a number on my resume.
I got stuck on evaluation design: if you evaluated an LLM for this service, what user types would you build and how would you analyze the results? The job description listed "building an LLM evaluation system for the domain" as the team's work. I had read that line without asking how I would do it myself, and my answers stayed unorganized.
Next time
- For every task in the job description, write down how I would design it. Interview questions come from there.
- Spend more time on design practice than on definitions. "How would you build X" mattered more than "what is X."
- Say when I do not know, but show the order I would think in. Forcing an answer made it worse.
Both stages said the same thing: the textbook path is not enough for the test, and concepts are not enough for the interview.