VK Interview Questions
& Process

They did a live coding session with 2 medium leetcode problems. Then, there was a deep dive on transformers, including training and inference. They also did some practical coding, like implementing an attention head. They discussed past projects with the team and lead, and assessed collaboration skills and role alignment. It was rigorous, practical, and fast-paced, looking for builders who can ship LLM features.

First, there was a technical quiz done in person at their office during work hours, and they insisted it couldn't be done remotely. However, after taking the time to commute and wait, it turned out no one actually monitored the quiz. A week later, I had a technical interview. The interviewers were friendly, and the questions weren't excessively difficult, though some were highly specific. They covered topics like the complexity of collection methods and the java.util.concurrent package, but avoided discussing JMM or different garbage collection methods. The most challenging question involved the memory representation of an Object type, which led into deeper hardware-related questions, making the overall interview quite demanding. They did correct some of my answers during the interview; for instance, they claimed hashCode() is cached by the JVM for all objects, including mutable ones, which I believe is incorrect. We also disagreed on other topics where my understanding seems to be more accurate.

My interview experience started with what the HR called a 'good' first interview. After that, I was given a time-consuming trial task which also got a positive response. Surprisingly, HR then invited me for a second interview, claiming it was to 'discuss my possibilities in the company'. In the end, I received no offer and no explanation, which left me with a very bad impression. It felt like I was tricked into wasting my time. The first interview went well, according to HR. They then presented a trial task to improve certain system features in production. Although I usually avoid trial tasks, I agreed due to professional interest. The trial task involved machine learning on a mixed dataset of text and images. There was no provided download server for the dataset, so I had to fetch images at a slow speed, and the text data was in a badly formatted CSV file that I had to fix. They also declined my request for server access for computations, suggesting they hadn't prepared the task well. Despite these issues, I successfully completed the task, improving their baseline recognition rate significantly and confirming their hypotheses, investing a lot of time. The second interview, which HR said was for 'discussion of my possibilities and trial task results review', surprised me because my trial task results were not discussed at all. This made me question why I was asked to do a long trial task instead of just a couple of short interviews. From the second interview, I realized Odnoklassniki wasn't looking for a Data Mining Engineer as advertised, but someone to implement their end-to-end ideas, which differs from the role description.

I was interviewed in 3 steps: first, I got a call from the company recruiter. After that, I waited literally a month for the next call, which was an invitation to visit the company for an interview. The interview process itself consisted of a meeting with HR and the head of the department (around 3 hours) and then a meeting with the vice-president (around 15 minutes). HR apparently forgot about me, which caused the month-long delay.