Scale AI SWE Intern Debugging Interview: 9 Experiences
- Company
- Scale AI
- Reports
- 9 interview reports
- Level
- Internship
- Stages
- OA → Phone screen → Onsite
Scale AI SWE intern interviews in fall 2026 centered on a debugging round; the 2024 loops used card and poker games built with classes.
At a glance
- Process: in fall 2026 the technical interview was a 60-minute debugging practical in Python or TypeScript, then a 30-minute Scale Credo (values) interview and a 30-minute hiring manager screen. In London (01), the other two rounds only counted if debugging went well. One candidate took an OA first.
- Asked recently: fixing three bugs in an unfamiliar codebase (01), a card game (03) and a poker hand (05) built with classes, a letter-frequency string problem after a Roman numerals swap (04), and two linked graph questions in an OA (06).
- Difficulty: candidate 02 called the behavioral part very easy; 03 didn't finish the card game and 05 said theirs went badly. One of the six current candidates got an offer (04).
- Practise first: Best Poker Category with a Joker, Sort Letters by Frequency and debugging projects in the Scale AI project coding bank.
Coverage: Software Engineering Intern / Software Engineer (Internship) candidates in San Francisco, Phoenix, London and the U.S. (city not given); one report gives no location. One online assessment, technical screens, onsite rounds and behavioral talks, May 2022 to October 2026. Two candidates got offers (04, 09). Entries 07–09 (2022) are more than three years old and sit under History, outside the current counts.
The intern loop right now (fall 2026)
Three Software Engineering Intern candidates described their Scale AI loop in fall 2026. All three had the same kind of technical interview: a debugging round. The candidate gets code they haven't seen before, with bugs in it, and has to find and fix them.
- October 2026, London. After the CV screen, this candidate booked three interviews together: debugging, behavioral, and hiring manager. The last two only counted if the debugging round went well. Their code had three bugs. The full account is entry 01 below.
- September to October 2026, OA first. Another candidate got the online assessment in September. By early October they had a debugging interview lined up for the same intern role. They didn't say what the OA asked.
- September 2026, a full onsite schedule. A third candidate shared their virtual onsite plan before going in:
- A 60-minute debugging practical, in Python or TypeScript.
- A 30-minute Scale Credo interview. The Credo is Scale's list of company values, so this is a behavioral talk built around them.
- A 30-minute hiring manager screen, on a second day.
The second and third candidates posted before their interviews and didn't say how they went, so they aren't counted in the tables below.
What came up
These counts cover the six current reports (01–06). Practical coding beat textbook puzzles: a debugging round, two card or poker games with classes, and only one standard string problem.
| Topic | Reports | Last seen |
|---|---|---|
| Card or poker game with classes (OOP) | 2 of 6 | Sep 2024 (report date) |
| Debugging an existing codebase | 1 of 6 | Oct 2026 (report date) |
| Questions about past projects | 1 of 6 | May 2025 (report date) |
| String problems | 1 of 6 | May 2024 (report date) |
| Graph questions (OA) | 1 of 6 | Oct 2023 (report date) |
Questions we found
Algorithms / Data Structures
| Question | Reports | Last seen |
|---|---|---|
| Reorder a string by letter frequency | 1: 04 | May 2024 (report date) |
| A Roman numerals problem (swapped out mid-interview) | 1: 04 | May 2024 (report date) |
Object-Oriented Coding (card games)
| Question | Reports | Last seen |
|---|---|---|
| Card game with a few simple functions | 1: 03 | Sep 2024 (report date) |
| Poker-hand question set up with classes | 1: 05 | spring 2024 |
Debugging
| Question | Reports | Last seen |
|---|---|---|
| Find and fix three bugs in a given codebase | 1: 01 | Oct 2026 (report date) |
Behavioral (BQ)
| Question | Reports | Last seen |
|---|---|---|
| Tell me about your past projects | 1: 02 | May 2025 (report date) |
6 candidate timelines, 2023–2026
01: London debugging round with a quiet interviewer
Report date: Oct 2026 · Role: Software Engineer (Internship) · Location: London, UK · Result: No offer
CV screen and scheduling
- They applied online. After the CV screen, they booked all three interviews at once: technical, behavioural and hiring manager.
- The behavioural and hiring manager rounds only count if the technical round goes well.
Technical interview (debugging)
- They got an existing codebase with three bugs in it and had to find and fix them.
- They found it confusing. The interviewer stayed quiet and gave little help.
02: easy behavioral talk, stopped at the technical round
Report date: May 2025 · Role: SWE Intern · Location: United States · Result: No offer
Rounds
- A behavioral interview and a technical interview. They did not say which came first.
- The behavioral part was very easy. They were asked about their past projects.
- They did not get past the technical interview and did not share its question.
They called the whole process casual.
03: card game with simple functions
Report date: Sep 2024 · Role: Software Engineering Intern · Location: San Francisco · Result: No offer
Technical interview (OOP)
- A standard object-oriented question: implement a card game with a few simple functions. They did not say which game.
- They stumbled and did not finish, but still called it very manageable and not much like LeetCode.
04: string by letter frequency after a Roman numerals swap
Report date: May 2024 · Role: SWE Intern · Location: United States · Result: Offer, accepted
Live coding (one round)
- The first problem was about Roman numerals. The candidate didn't know how Roman numerals work, so the interviewer switched to a different problem.
- The second problem: reorder a string based on how often each letter appears.
- The interviewer said plainly that the question was not hard.
That single live coding round was the whole technical part they described, and it ended in an offer.
05: poker-hand question, first technical interview ever
Interview date: around spring 2024 (candidate's estimate; posted Jun 2026) · Role: Software Engineer (Internship) · Location: San Francisco · Result: No offer
Virtual technical interview
- A poker-hand question like the simple poker problems on LeetCode, but set up with classes.
- It was their first technical interview. Nerves got the better of them and it went badly.
- The interviewer was very friendly and supportive the whole way.
They applied online, and the full process took about five months.
06: graph OA, then the intern spots filled
Report date: Oct 2023 · Role: Software Engineer Intern · Location: United States · Result: No offer
Online assessment
- Two closely related graph questions, with two weeks to finish.
Technical phone interview
- Booked, then canceled. The internship spots had filled up.
History (2022)
These three reports are more than three years old, so they stay out of the counts above. Their named questions:
| Question | Reports | Last seen |
|---|---|---|
| Assign unit-time tasks to time slots, one task per second | 1: 08 | Sep 2022 (report date) |
| Tell me about your past experience | 1: 08 | Sep 2022 (report date) |
| Debug a Python framework | 1: 09 | May 2022 (report date) |
07: a design question with an algorithm inside, stopped at the first technical
Report date: Sep 2022 · Role: SWE Intern · Result: No offer
Process
- The recruiter sends a slide deck that lays out the steps: recruiter phone screen, one technical round, then a final onsite of three interviews.
First technical interview
- A tricky question with a high bar. It was more software design than algorithms, with an algorithmic part, and nothing like LeetCode. They did not share the prompt.
- They did not pass, so they never reached the onsite.
08: one scheduling question, then past experience
Report date: Sep 2022 · Role: SWE Intern · Location: Phoenix, AZ · Result: No offer
Interview
- One coding question about a job schedule: tasks get assigned every second, and each task takes exactly one unit of time. They gave no more detail.
- Then questions about their past industry experience.
09: two phone screens, then practical, debugging and behavioral
Report date: May 2022 · Role: Software Engineering Intern · Location: United States · Result: Offer, declined
Technical phone screens
- Two technical phone screens, one hour each.
Onsite
- A coding interview on the practical side.
- A debugging interview: they had to debug a Python framework.
- A behavioral interview.
One more intern report (role not named)
This candidate did not say which intern role they interviewed for, so the report is not counted above.
Intern debugging round across many files
Interview date: summer 2026 (year taken from the post date) · Role: Intern (role not named)
Debugging interview
- The code to debug was spread over several files, with layer after layer of nested functions and classes.
- The same candidate later did Scale AI's new grad SWE debugging round and found this intern round the harder of the two.
Patterns across the reports
-
Card and poker games built with classes. Entries 03 and 05, both San Francisco in 2024, got a card-game style question in an object-oriented setup. One was a card game with simple functions; the other was a poker-hand variant. Entry 03 did not finish and entry 05 said it went badly, yet both described the interview as manageable or friendly.
-
Debugging someone else's code. Entries 01 and 09 had a debugging round, four years apart. In 2022 it was one onsite round among three, on a Python framework. In 2026 in London it was the only technical round, and it decided whether the other two rounds counted.
-
Talking about past work. Entries 02 and 08 were asked about past projects or past experience. Entry 02 called that part very easy; entry 08 did not say. Entry 09 also had a behavioral round but did not share the questions.
-
One technical round as the gate. In entries 01, 04 and 07, a single technical round decided what happened next. In London it decided whether the behavioral and hiring manager rounds counted, in 2024 it was the only coding round before an offer, and in 2022 failing it ended the process before the onsite.
Common questions
What is the Scale AI debugging interview?
You get code you haven't seen, with bugs in it, and have to find and fix them. Candidate 01's had three bugs; the uncounted intern report below describes code spread over several files with nested functions and classes. In fall 2026 it ran 60 minutes in Python or TypeScript.
Does Scale AI give interns an online assessment?
Sometimes. Candidate 06 had two linked graph questions with two weeks to finish, and one fall-2026 candidate got an OA before the debugging interview.
What is the Scale Credo interview?
A 30-minute behavioral talk built around Scale's company values (the Credo), part of the fall-2026 onsite schedule.
Keep practicing
- Algorithms: practice versions of the reported questions: Sort Letters by Frequency and Roman Numeral to Integer (entry 04), Best Poker Category with a Joker (entry 05), and Schedule Unit-Time Jobs for the Most Profit (entry 08). More Scale AI coding questions are in the algorithm question bank.
- Debugging and practical projects: the Scale AI project coding bank has Scale AI's repository-style exercises, including debugging projects.
- Everything Scale AI in one place: the Scale AI interview page lists Scale AI's coding questions by OA, phone and onsite stage, plus its design and project exercises.
- More firsthand reports: browse the Experience feed.
Sources: Anywhere useful interview info can be found.
Comments
0No comments yet — be the first to share what you know.