I prepared for the VO for a long time, and I still ended up getting an add-on interview — I can accept that outcome. I'm hoping someone kind who's been through an add-on interview can share the question bank!!
Will they pull from question types that come up a lot but I wasn't asked, or will they pull from really old questions? I've read a lot of other add-on interview posts and haven't gotten anything concrete out of them. I get it, though — people who already went through it probably don't come back to the forum. If anyone's recently gotten an add-on interview, feel free to reach out, let's prep together.
Below is my VO experience — giving back to the community.
The three rounds, in order, were: short video (Reel), car rental, and carpool.
When I was prepping, I assumed car rental and carpool were part of the same question set — turns out they're not. Also, I think there might really be something to how the question combos get grouped. My experience was pretty close to a post I found on here.
So my suggestion is to look at the ordering/combos in other people's posts — that's a prep method in itself.
Product Sense
Don't spend too much time on this part — not in prep, and not in the interview either. Just get the questions from other people's interview reports down cold. The more you prep PS, the more likely you are to freeze up, ironically.
For example, "how would you measure whether xxx is successful" — you can go into a lot of detail on this, but if you get too deep into the weeds it's hard to explain clearly, and you might not even hit what the interviewer is looking for. If the interviewer doesn't follow what you're saying they'll keep asking follow-ups, which wastes time. Keep in mind this section is only 10 minutes, so they're not going to go that deep — focus on the main points.
When you're reviewing interview reports to prep, I'd suggest working backwards from the SQL and Python questions: figure out what SQL and Python asked for that topic, then you can casually bring that up when you're prepping Product Sense.
Data Modeling
Same thing — don't spend too much time here either. They won't have you model the whole product, it's basically just one core piece of the business. In the interview, the prompt will be pretty detailed and will tell you directly which part to focus your modeling on.
So again, prioritize. The core of this section is: what entities are there, what are the relationships between them, what's the key, and give the core fields for each entity. Be clear about which tables are dim tables and which are fct tables, and don't forget to give some key fields for the dim tables too.
SQL
There's not a ton to say about this part, but this is exactly where I got flagged. When you're nervous in the interview, it's easy to not fully understand the question and to fumble some of the details.
The questions aren't hard. Practice the ones from other people's interview reports a lot, and talk it through with ChatGPT a lot — have it check whether your answer is concise and clean enough.
Python
This part is entirely questions straight from other people's interview reports. Practice a lot — you need to be able to write it fluently, because there's a good chance you'll spend too much time on the earlier sections and end up with no time left for Python.
You absolutely have to get each question down to under 7 minutes. In both my first and second rounds I spent too much time on the earlier parts, so I only had 7 minutes left for Python. Good thing I was fluent enough — otherwise I wouldn't have even gotten the add-on interview.
During the interview, the interviewer will have you write your answer down (for every section). What you write down is what determines whether you pass. No matter how well you explain something verbally, the interviewer will forget it by the time they're writing up feedback afterward — what you wrote is the evidence of whether you pass.
My guess is a lot of people feel like they interviewed well but get bad feedback because what they actually wrote down wasn't great.
Now let me share my thoughts on each round.
Reel
This round I got the "leads you on but never commits" type of interviewer — showed up late, ignored me, hard to get any back-and-forth with, multitasking the whole time. Basically neither proactive nor a flat-out no. This was also the interviewer who gave me the worst feedback of the three.
But honestly I didn't perform that well either — it was my first round, I was really nervous, and I was still feeling my way through a lot of the steps.
Right at the start I told him I knew this interview was time-sensitive and asked if he could help me keep track of time for each section. He said yes verbally but never actually did it. Then in the Python section he told me there were only 7 minutes left, and the question had a decent amount of code to write (or maybe my answer just wasn't concise enough).
I'd written more than half of it when he had me stop, saying he understood my approach and time was up. I looked at the clock and it was exactly on the hour, but he'd started late, so I immediately told him — could I finish writing, since we'd started late in the first place? He let me write for one more minute after that, said he really had somewhere else to be, and promised he'd pass me on this question. My final feedback said 90% correct — so it did count as a pass. (The Python question was a buffer problem, straight from other people's interview reports — those posts are enough to cover it.)
Then I got flagged on SQL — supposedly something wrong with the join in the first question. I honestly don't remember the question anymore, so I don't know exactly what the issue was. Sorry I can't be more helpful there.
But during the interview itself, the interviewer never pointed out any problem with my query — it was all "ok", "good", that kind of feedback. I did write it pretty haltingly, though (writing queries in a plain text doc feels really awkward to me, and the question was long, so I had to keep scrolling back up to find table names — it was a pain).
So one thing's for sure: the interviewer didn't fully understand my answer in the moment, and only caught the problem later during the write-up.
PS and DM were the standard questions — evaluating short video's impact on the newsfeed. This being my first round, I was insanely nervous, could barely get my words out, and my hands were shaking while typing. The focus was around user, content, and the interaction between user and content — how one piece of content can include text, images, and video.
The DM follow-up was about share — other people's posts cover this too.
Car Rental
I won't go over the questions again since they matched the post I linked above. This one was with an interviewer who was very careful, had great back-and-forth, and really took the time to understand my answers.
At one point I thought I'd bombed this round. I could only find one interview report on this topic, and I even assumed it was the same topic as carpool, so I basically didn't prep for it — I figured it wouldn't come up. It wasn't until half an hour before the interview that something felt off, so I chatted with ChatGPT about PS and DM for it. Then it actually did come up and I was ready to give up.
Luckily, GPT's answers were pretty good, and I basically just went with what GPT said — I hadn't thought deeply about this question myself. It turned out fine, though. The interviewer would follow up, but with the experience from my first round, I had a decent read on what interviewers actually care about. PS and DM both went smoothly.
For this topic I got stuck on SQL again. I hadn't prepped any questions for this section, so I was basically doing it live. They gave me 4 tables: user, car, location, fct_rental. There were 2 questions; I only remember the second one because I got stuck on it for a long time.
Roughly, as of the end of a given day, calculate how many cars of each car size were rented out in each city, and how much inventory that city has for that car size, then work out the ratio (basically rented cars divided by inventory). Overall you need to group by (location id + car size). Where I got stuck was figuring out how to determine whether a car had been rented out as of a given day. I'm still not sure whether I got it right in the end.
This interviewer was pretty thorough — they pointed out every single issue in my SQL. That ate up a ton of time, so once again I only had 7 minutes left for Python.
Luckily I'm fast at Python, so I got through everything in the end. The Python question was identical to what was in the post — a straight repeat.
At one point I thought I was going to fail this round on SQL, but this interviewer must have been kind about it, because they didn't fail me. Quite the contrast with the flaky interviewer from round one.
Carpool
The easiest round of the three. With the experience from the first two rounds, I didn't really run into any difficulty in this one. I answered everything fast, accurately, and decisively.
SQL was the easiest — didn't even need a CTE, see the post I linked above, straight from other people's interview reports. Python was also a repeat — a sweep-line problem, very simple. This is the kind of question that's dead easy once you know the approach.
I finished all the questions in 50 minutes. This was the easiest topic to prep out of all of them.
Anyway, I'm genuinely grateful to get one more shot at this. Once again, I'm begging anyone with experience to share what the add-on interview is like. Please, please, please.
Discussion
Loading comments…