My background:

• 4+ YOE across Data roles
• Currently Staff DS at a startup

Round 1: DSA(Dec 2025)

Variation of 726. Number of Atoms where you have to calculate the weight of the molecule given a fixed mapping of the weight of each atom. Got the base test cases down but was failing on some, manage to talk through the implementation that would work. Received an email the next day to schedule System Design round.

Round 2: System Design(Jan 2026)

First interviewer didn't show up the first time, had to reschedule.

First 15 mins was talking about a project that you are proud of.
Remaining 45 mins was about System Design.
This question was not a traditional System Design round as what their HR told me. Was more of a Data Engineering design where you had a supermarket system to track SKUs of items in stock across 100s of stores.

There are two streams:

• stock, items coming in at ~1M QPS(item_id, number_of_items either +/- indicating buy/sell, quantity_before, timestamp) from stores.
• camera, contains images of items on shelfs taken everytime there are changes on the store ~100K QPS.

You had a database with metadata about the items and stores. The question is how to generate some signals that are given from these data which are features fed to a downstream ML model. You were given some signals that had to be calculated using states of items and joined with the metadata. Some were tenporal in nature. The catch was you didnt have to calculate features for each item event. Just store them in Redis/DB and then build the inputs on eahc camera event.

Focus was on:

• scale
• data quality
• handling deduplication

Question completely threw me off, seemed too domain specific and I had to spend a lot of time understanding how the items and camera signals are coming. Didn't have time to solve data quality and thought process was not coherent.

Received rejection a few days later.