RepostedScale AI logo

Scale AI · OA → Phone screen → Onsite

In process

Scale AI ML Research Engineering Intern Interviews: 1 Experience

Machine learning engineer · InternshipJan 2025
Shared byjanebeingjane

A Scale AI ML research engineering intern candidate in early 2025 had a 6-hour paper-implementation OA, then coding on a top-k sampler and attention block.

At a glance

  • Process: 6-hour HackerRank OA → recruiter contact → three remote rounds (ML coding, ML concepts, behavioral), January to March 2025 (report 01).
  • Asked: a top-k sampler and a multi-head attention block in the coding round; post-training, fine-tuning, reinforcement learning and LoRA in the concepts round (report 01).
  • Difficulty: the candidate did not finish the OA but still got interviews; they passed the coding round the same day, then were rejected after the final two rounds.
  • Practise first: Top-K Sampler and Multi-Head Attention Block.

Coverage: 1 report · Machine Learning Research Engineering Intern (Summer 2025) · January–March 2025 · location not stated.

What came up

The live rounds were all deep learning: two building blocks to code, then post-training and fine-tuning theory. Classic ML algorithms were not asked. The OA was a research task: implement an algorithm from a paper.

Topic Reports Last seen
Deep-learning building blocks in code 1 of 1 Mar 2025
Post-training, fine-tuning, RL and LoRA 1 of 1 Mar 2025
Implement a paper's algorithm (OA) 1 of 1 Jan 2025

Questions we found

ML coding

Question Reports Last seen
Top-K Sampler 1 Mar 2025
Multi-Head Attention Block 1 Mar 2025

ML concepts

Question Reports Last seen
Post-training and fine-tuning, with RL details and LoRA 1 Mar 2025

Online assessment (OA)

Question Reports Last seen
Implement an algorithm from a research paper (paper not named) 1 Jan 2025

1 detailed candidate timeline

01: from an unsubmitted OA to three remote rounds

Report date: March 21, 2025 · Role: Machine Learning Research Engineering Intern (Summer 2025) · Result: Rejected

OA (received January 27)

  1. A 6-hour HackerRank assessment: implement an algorithm from a research paper. The candidate did not name the paper but found the task fun.
  2. They did not finish or submit it, and HR still reached out to schedule interviews.

Recruiter contact (February 5)

  1. A recruiter set up the interviews.

Round 1: ML coding (March 4)

  1. Write a top-k sampler, then a multi-head attention block.
  2. No classic ML (traditional algorithms rather than deep learning) was asked.
  3. The interview was in the morning; by the afternoon the candidate heard they had passed to two more rounds.

Round 2: ML concepts (March 11)

  1. Many questions on post-training (training done after a model's initial pretraining) and fine-tuning.
  2. Reinforcement learning details; the candidate gave a rough outline and said they did not know the finer points.
  3. How to fine-tune with LoRA (Low-Rank Adaptation: train small added weight matrices while the original weights stay frozen); they answered these fully.

Round 3: behavioral (March 12)

  1. A relaxed talk; the candidate told four or five short stories (prompts not shared). The interviewer said they were very pleased with the conversation.

Result (March 21)

  1. Rejected; the email called the process a remote onsite.

Common questions

What is the Scale AI ML research intern OA?

In this report, a 6-hour HackerRank task to implement an algorithm from a research paper (report 01).

What coding does Scale AI ask ML research interns?

Deep-learning building blocks: a top-k sampler and a multi-head attention block (report 01).

What ML concepts does Scale AI ask?

Post-training and fine-tuning, reinforcement learning details, and fine-tuning with LoRA (report 01).

Keep practicing

Sources: Anywhere useful interview info can be found.

0 total reactions. Your reactions: none.
0 comments

Comments

0

No comments yet — be the first to share what you know.