Hi all! I got an offer from Snowflake for the IC1 level, and I'm posting my loop here in case it helps someone who's getting ready.
It began with a recruiter contacting me, followed by a first conversation with the hiring manager. We went over the team, when the interviews would happen, and what they were looking for. Nothing surprising, but it gave me some context before the real interviews.
Next were two DSA screens, and I'll admit they rattled me. The questions were out of the ordinary, but the harder part was that Snowflake wants you to invent your own test cases, including the nastiest edge cases, and run them live to prove your code works. Normally I'd write a solution and try the sample inputs I was given. In these screens the checking was entirely on me. I had to think about correctness like a QA engineer reviewing my own work, and that change of mindset was what I remember most from the screens.
I passed both, and the onsite had three rounds: DSA, system design, and behavioral. The behavioral one was fairly typical, with topics like ownership, ambiguity, deep dives, and conflicts.
DSA, onsite
To prepare, I'd already finished a well-known playlist of standard problems organized by pattern, so I put most of my remaining time into LeetCode questions tagged with Snowflake. Those turned out to be a very good preview of the real thing.
My question started out as a web crawler, grew into a multi-threaded crawler, and ended up as a distributed systems discussion, all inside one round. Each follow-up peeled back another layer. A few of the harder ones:
- The function that gives you a node's neighbors is hosted on a remote server. What happens if the network gets partitioned?
- Say there are 1B nodes and the BFS branches out massively. How would you make building the shortest path faster?
- If a billion-node graph has lots of possible paths, what further optimizations are there?
- When several threads build paths in parallel, how do you handle races on shared state?
I also noticed the interviewers took their follow-ups from my own suggestions. I brought up multi-threading to speed things up, and that directly produced the race condition question. Think before you propose something, since you'll be asked to justify it and push it further.
System Design, onsite
I was asked to design a photo album. That seems easy, yet the requirements changed several times as we went. Adjusting your design mid-conversation is a big part of it, so don't cling to your first idea.
At IC1 level you're basically expected to drive the discussion in the first half, and in the second half the interviewer probes the design. Another thing to know is that Snowflake sticks to infrastructure questions on the backend side rather than product-style prompts such as designing WhatsApp.
Follow-ups I got:
- When an image is uploaded, how do you keep the metadata DB and the blob storage consistent with each other?
- The user wants to see their photo right after uploading it, even though background processing may be behind. How do you handle that?
- How would you block one user from uploading the identical photo repeatedly?
- Why did you pick Postgres? When would a relational database start to struggle, and what would you switch to?
Once those deep dives were done, we covered the schema, the APIs, and the functional and non-functional requirements. The round felt very complete.
Final thoughts
Snowflake's process struck me as one that rewards correctness and depth. Solving the problem is not enough; they want evidence that you thought it all the way through. The self-written test cases in the DSA screens set them apart from other companies I've seen. If you're preparing, do more than grind LeetCode: get used to writing tests yourself, check edge cases before writing code, and talk through your tradeoffs aloud.
Feel free to ask me anything, and best of luck to everyone preparing!
Discussion
Loading comments…