Scale AI Product/Software Engineer Onsite Interview Experience

Round 1 — BQ + Backend Practical Behavioral questions covered previous projects. For the backend practical the stack chosen was Python. The task was similar to implementing a lightweight load balancer

Round 1 — BQ + Backend Practical Behavioral questions covered previous projects. For the backend practical the stack chosen was Python. The task was similar to implementing a lightweight load balancer; you had to clone a specified project from GitHub. Required implementation points included: a. Worker state management (e.g. active/overloaded/unreachable) b. Task queue and priority scheduling mechanism c. How to support scalability (for example, dynamic joining of worker nodes) You needed to design from perspectives such as task-dispatch logic, worker heartbeat mechanism, failover strategy, etc., and produce code within a limited time — the bar was fairly high. --- Round 2 — Coding: Clock-hand angle The coding round was done live on CoderPad and you had to explain your approach. The problem itself is not hard but you must handle edge cases and clearly explain the logic. Given a time string like “3:45”, compute the angle between the hour hand and the minute hand. You can apply a direct formula, but the interviewer wanted you to explain where each part of the angle calculation comes from. The idea: the minute hand moves 6° per minute; the hour hand moves 30° per hour plus 0.5° per minute. So compute each hand’s angle relative to 12 o’clock from the input hour and minute, take the absolute difference, and if it’s greater than 180° subtract it from 360° to get the smaller angle. Follow-up: how would you adjust the formula if the input also includes seconds or milliseconds? --- Round 3 — System Design This round was a training session: a smiling Japanese interviewer shadowed a Chinese candidate. The task was to design a Ticketmaster-like system. You can follow Alex Xu’s methodology; however, the candidate said: “Let’s not waste time on back-of-the-envelope calculations — forget about distributed systems. Let’s map the user flow clearly and draw a diagram with all components.” Requirements included: a. How to handle flash-sale scenarios where many users try to buy tickets in a short time b. How to implement a timeout on the purchase page and what to do if payment is not completed within the timeout c. How to handle the situation when tickets are sold out d. How to guarantee that users who completed payment will definitely receive tickets e. How to implement a waitlist that notifies the next-in-line people when tickets are returned --- Round 4 — BQ + Coding Behavioral questions were standard: 1. “Tell me about a time you had to learn something quickly.” Show structured learning ability — e.g., using API docs, codebase walkthroughs, shadowing colleagues, building a proof of concept (POC). 2. “Tell me about a time you disagreed with a teammate or manager.” Support your stance with data-driven or customer-impact reasons, show you can accept valid feedback, and turn the result into a win-win. 3. “Tell me about a challenging project you worked on.” Use the STAR structure to tell a business-impactful story; in the Action section quantify your measures (for example, “improved performance by 40%,” “reduced latency to P99 < 200ms”). The coding problem: given a tree where you only know each node’s list of children, find the lowest common ancestor (LCA) of two nodes. Several approaches were discussed: first, run DFS to compute each node’s parent and depth, then raise the deeper node upward until depths match and move both up until they meet. Another approach: run a DFS from the root where each node returns a pair of booleans indicating whether it can reach the two target nodes; the first node that returns (true, true) is the LCA. Ideas, clarifications, comments, and test cases were all explained in detail. Other Scale AI interview experiences you can refer to this.