Company interview guide

Runway Machine Learning Interview Guide (2026)

Short, practical coding and video-centric ML design screens, a code review of your own work, then culture rounds and calls with the founders.

Runway's machine learning interview loop has 7 reported stages, from recruiter call to leadership calls; typical length: 2-5 weeks.

Roles covered: ML Engineer, Research Scientist and Applied Scientist. The question bank holds 3 questions reported from Runway interviews.

Try a free question from this loop's topics

Interview process length
2-5 weeks
Stages
7
Reports linked to bank questions
5
Bank report dates
Feb 2022–Aug 2025
Questions in bank
3 (0 free)
Last reviewed
October 2026

How the Runway interview loop works

A recruiter call, then usually two technical screens of 30-45 minutes each: a coding round (often job-relevant Python or a small ML algorithm) and a system design round. Some research candidates instead do a code review of their own project and a research discussion of a paper. Next comes a virtual onsite of about four or five short interviews over two to three hours, mixing a technical deep dive, more design or a PyTorch debugging round, and culture interviews. The loop closes with short calls with leaders such as the CEO, CTO or CPO. Most 2024-2026 reports finish in two to three weeks and get feedback the same day; a few older reports took four to five weeks. The company's job pages do not describe the stages, so this outline rests on candidate reports only.

The Runway interview process, stage by stage

01Recruiter call

About 30 min with a recruiter

A conversation about your background, why Runway and practical details such as location, relocation and compensation. Candidates are sometimes redirected to a different opening than the one they applied for. One candidate was turned away after being asked what they wanted to know about Runway, so arrive with informed questions about its models and research.

What it tests

  • Why Runway, and why generative video or world models
  • How your experience maps to the team
  • Logistics: location, relocation, level and compensation

How to prepare

  • Have specific questions ready about Runway's recent models and products; you may be asked what you want to know.
  • Ask which rounds follow and what each covers; candidates say the recruiters explain this clearly.
  • Mention the team you most want; the same title covers research, platform and product work.

A recruiter or HR call opens the process in 7 ML and research reports and 5 engineering-role reports; it is a shared stage, so all are counted.

02Coding screen

30-45 min live coding, usually Python; a few teams also use TypeScript or a code sample review

A short, practical coding round rather than a puzzle. ML candidates report implementing a small ML algorithm in numpy or PyTorch, an n-gram language model with fit and generate methods, or a classic data-structure task such as an LRU cache. Engineering-track reports describe a concurrency lock, finishing a partial Python implementation and adding tests. Time is tight, so follow-up parts are often only discussed.

What it tests

  • Writing correct, readable Python quickly
  • Implementing a basic ML algorithm or language model from scratch
  • Probabilistic text generation from counts, and choosing model order
  • Testing and edge cases in a small implementation

How to prepare

  • Practise building small models from scratch in numpy within 30 minutes.
  • Aim for a working core first; follow-ups such as choosing n for an n-gram model may only be discussed.
  • Write a couple of quick tests as you go; one interviewer asked for them explicitly.

A coding round in 12 ML and research reports, 6 since 2024, usually the first technical screen; 8 engineering-role reports describe a similar round but are not counted.

03System design screen

30-45 min open discussion; often a second screen, sometimes repeated at the onsite

The most consistent technical round. For ML and research roles the prompt is very often a text-to-video or video-to-video search system, covering both the models and the serving side; one older loop asked how to pick clips for a movie trailer. Engineering roles get prompts built around Runway features, such as handling large media assets, GPU-backed generation jobs or a browser video editor. Several candidates found 30 minutes very short and were pressed with rapid follow-ups.

What it tests

  • Multimodal embeddings for video and text, and how to segment video
  • Indexing and approximate nearest neighbour search at large scale
  • Ranking, freshness, latency and cost trade-offs
  • Model choices versus infrastructure choices, depending on the interviewer
  • Designing around GPU-heavy generative workloads

How to prepare

  • Prepare a video retrieval design you can present in ten minutes, then use the rest for follow-ups.
  • Skip long requirement gathering; state assumptions quickly and move on.
  • Find out the interviewer's background early and balance modelling detail against infrastructure.

A system design round in 12 ML and research reports, 6 since 2024; 6 of them name a video search or retrieval prompt. Engineering-role reports describe feature-based designs.

04Code review & research discussion

About 30 min each; mainly research-leaning candidates, sometimes in place of a live coding screen

Instead of, or alongside, live coding, some candidates pick code they wrote, often a public repository, and walk the interviewer through its design and implementation. Research candidates may also discuss a paper of their choice, with questions on ML practice such as debugging training runs. Engineering loops have a related round where you review someone else's code.

What it tests

  • Explaining the structure and trade-offs of your own code
  • Code quality, maintainability and testing habits
  • Reading and critiquing a research paper
  • Practical ML judgement, such as diagnosing a failing training run

How to prepare

  • Choose a repository you know line by line and can defend; clean it up beforehand.
  • Pick a paper close to video generation or world models and prepare what you would change.
  • Expect questions on what went wrong and how you found it.

A review of the candidate's own code in 4 ML and research reports since 2024, two with a paper discussion; one more report of unclear role and 3 engineering reports describe code review rounds.

05Virtual onsite: technical depth

Part of a 2-3 hour virtual onsite of about four or five short interviews

The onsite technical rounds go deep on your area. Reports describe a deep dive into past research or projects with a senior research scientist, an ML depth interview, a PyTorch debugging round, a walkthrough of a system you designed, and estimating the compute and memory needs of a neural network. Some onsites add another design or coding round.

What it tests

  • Depth behind the projects on your resume
  • Finding bugs in PyTorch training code
  • Estimating parameters, FLOPs and memory for a network
  • ML performance and systems knowledge in your specialty

How to prepare

  • Practise counting parameters and activation memory for common architectures by hand.
  • Rehearse debugging a broken PyTorch training loop: shapes, masks, loss and optimizer steps.
  • Prepare one project to explain end to end, including the numbers.

Technical depth, deep-dive or debugging rounds after the screens in 8 ML and research reports, 4 since 2024; the content varies by team.

06Culture & behavioral interviews

1-2 short conversations in the onsite, often with a people-team member or engineering manager

Candidates say culture fit carries real weight. Expect questions on motivation, why you are leaving your current role, delivering something quickly and the hardest technical problem you have solved. Several candidates describe these as relaxed conversations about what you want to work on.

What it tests

  • Motivation for Runway and for creative AI tools
  • Shipping quickly and working with ambiguity
  • Collaboration across research, product and design

How to prepare

  • Prepare stories about shipping fast and about tackling something others thought was not possible.
  • Be clear about what you want to build next; interviewers ask about it.

Culture or behavioral rounds in 7 ML and research reports and 6 engineering-role reports; this stage is shared across tracks, so all are counted.

07Leadership calls

Short calls with leaders, for example 15-30 min each with the CEO, CTO or CPO

The final step in most recent reports. The talks are informal and cover higher-level technical ideas, your past experience and what you would do at Runway. Some candidates met several executives in one stage.

What it tests

  • Clear, high-level thinking about your field
  • Ambition and fit with the company's direction
  • Your own questions about strategy and research

How to prepare

  • Read Runway's recent research and product posts so you can talk about where the company is going.
  • Bring a few thoughtful questions for each leader.

Calls with founders or senior leaders end the loop in 6 ML and research reports and 8 engineering-role reports; a shared fit stage, so all are counted.

Questions and themes reported most often

  • Design a text-to-video or video-to-video search system (6 reports, system design screen)
  • Walk through code you wrote, such as a public repository (4 reports, code review & research discussion)
  • Deep dive into a past research project (4 reports, virtual onsite: technical depth)
  • Implement a basic ML algorithm in numpy or PyTorch (4 reports, coding screen)
  • Discuss a research paper of your choice (2 reports, code review & research discussion)
  • N-gram language model with fit, generate and a choice of n (1 reports, coding screen)
  • Pick clips from a movie for its trailer (1 reports, system design screen)
  • LRU cache (1 reports, coding screen)

What Runway asks about

  • ML / system design30%
  • Coding & ML coding25%
  • Code review & past work20%
  • Culture & leadership15%
  • ML depth & debugging10%

Runway interview prep checklist

  • Can design a text-to-video search system covering embeddings, indexing, ranking and serving in 30 minutes
  • Built an n-gram language model with sampling and discussed how to pick n
  • Picked and rehearsed a repository of your own to walk through
  • Prepared a paper to discuss and a past project to deep dive on
  • Can estimate parameters and memory for a transformer or convolutional network by hand
  • Have specific questions about Runway's models for the recruiter and leaders

Tips

  • Make a video retrieval design your first priority; it is the most reported ML prompt.
  • Practise 30-minute versions of every round: a design you can sketch in ten minutes and code you can finish in twenty.
  • Have a clean repository of your own ready to present in case your loop uses a code review.
  • Drill small from-scratch implementations in numpy and PyTorch, plus debugging broken training code.
  • Learn how video generation and multimodal retrieval models work at a high level, since prompts are built around Runway's products.
  • Ask the recruiter which rounds you will have; the same title covers research, platform and product loops.
  • Expect fast decisions, often the same day, and keep your schedule open to move quickly.

Practice the Runway loop

Work the stages in order: every reported question for this company, grouped by round, with solutions in the app.

Runway interview FAQ

How many interview rounds does Runway have for ML roles?

Candidates report 7 stages: Recruiter call, Coding screen, System design screen, Code review & research discussion, Virtual onsite: technical depth, Culture & behavioral interviews and Leadership calls. A recruiter call, then usually two technical screens of 30-45 minutes each: a coding round (often job-relevant Python or a small ML algorithm) and a system design round. Some research candidates instead do a code review of their own project and a research discussion of a paper. Next comes a virtual onsite of about four or five short interviews over two to three hours, mixing a technical deep dive, more design or a PyTorch debugging round, and culture interviews. The loop closes with short calls with leaders such as the CEO, CTO or CPO. Most 2024-2026 reports finish in two to three weeks and get feedback the same day; a few older reports took four to five weeks. The company's job pages do not describe the stages, so this outline rests on candidate reports only.

How long does the Runway ML interview process take?

Reported timelines run 2-5 weeks.

What does Runway ask in ML interviews?

By reported question mix: ML / system design 30%; Coding & ML coding 25%; Code review & past work 20%; Culture & leadership 15%; ML depth & debugging 10%. The most repeated themes: Design a text-to-video or video-to-video search system; Walk through code you wrote, such as a public repository; Deep dive into a past research project.

How should I prepare for the Runway ML interview?

Make a video retrieval design your first priority; it is the most reported ML prompt. Practise 30-minute versions of every round: a design you can sketch in ten minutes and code you can finish in twenty. Have a clean repository of your own ready to present in case your loop uses a code review.

Are there free Runway interview questions?

LeetML has 3 questions reported from Runway interviews. The stage sections above link the free practice problems for the topics each round covers; the reported-question set itself is in the app.

Related interview guides

Official sources