Company interview guide

Nuro Machine Learning Interview Guide (2026)

Algorithm coding, hand-written ML building blocks and a deep dive on your own research, for an autonomous driving company.

Nuro's machine learning interview loop has 8 reported stages, from recruiter call to behavioral and leadership chat; typical length: Not well reported; few candidates give a timeline.

Roles covered: ML Engineer, Research Scientist and Data Scientist. The question bank holds 2 questions reported from Nuro interviews, 1 of them free to solve in your browser.

Try a free question

Interview process length
Not well reported; few candidates give a timeline
Stages
8
Reports linked to bank questions
2
Bank report dates
Jan 2026–Jan 2026
Questions in bank
2 (1 free)
Last reviewed
October 2026

How the Nuro interview loop works

Most reports start with a recruiter call (about 30 minutes in one research report) and then two technical screens, sometimes back to back. The screens pair a coding round with a second round that varies: ML coding in Python, an ML or ML-infrastructure design, applied ML questions, or a talk about your research. Data science candidates in 2021 got an online or overnight SQL and Python test first. The onsite is described as six to eight one-on-one interviews: one or two coding rounds, ML coding, research or domain rounds tied to the team, a research talk for research scientists and a behavioral round; one 2019 onsite included a chat with a co-founder. One 2024 research candidate was told each interview's topic in advance. Several candidates say they never heard back after a round, so follow up and keep other processes moving.

The Nuro interview process, stage by stage

01Recruiter call

About 30 min by phone or video

A conversation about your background, the role, logistics such as visa needs, and why Nuro. In one 2025 research report the recruiter walked through the next rounds. Ask here what each screen will cover, but be ready for the content to differ from the description.

What it tests

  • Why Nuro and why autonomous driving
  • How your past projects fit the team
  • Logistics: degree, location, visa, timeline

How to prepare

  • Ask the type, language and length of each screen; one 2025 candidate was told coding would be applied and then got a generic algorithm problem.
  • Have a short answer ready on why autonomy and why Nuro specifically.

Recruiter call as the first step in 6 ML and research reports (four since 2025); one research candidate describes a 30-minute call on background, past projects and motivation.

02Online test (data science, some software roles)

Online coding test or short-deadline assignment, SQL and Python for data science; one medium algorithm problem for a software role

Not part of the ML engineer or research loops in our reports. Data science candidates in 2021 described an online test with two SQL questions and one Python question, or an overnight assignment on SQL queries and implementing SQL-style logic in Python. A 2025 AI platform software candidate got a single medium algorithm problem.

What it tests

  • Writing correct SQL joins and aggregations against given tables
  • Reproducing SQL operations in plain Python
  • Solving a medium algorithm problem under time pressure

How to prepare

  • Data science applicants should drill SQL joins, grouping and window functions before applying.
  • If the assignment arrives with a tight deadline, ask for a reasonable window; one candidate got an evening-to-morning turnaround.

Two 2021 data science reports describe an online SQL and Python test or overnight assignment; one 2025 software report had an online algorithm test. No ML or research report mentions one.

03Coding rounds (screen and onsite)

About 45-60 min of live coding by video or shared doc, usually Python; C++ in some ML rounds; one or two rounds across the loop

The most consistent round. Most problems are generic medium-difficulty algorithm questions with little link to driving: an N-ary tree maximum path sum in 2026, an intervals problem in 2025, a path-sum dynamic programming variant and quicksort in older reports. A 2025 ML engineer screen instead focused on object-oriented design in C++. Several candidates say the prompt was unclear or poorly worded.

What it tests

  • Tree recursion and dynamic programming on medium problems
  • Clean object-oriented code, sometimes in C++
  • Clarifying an under-specified prompt before coding
  • Live SQL for data science candidates

How to prepare

  • Restate the problem and walk through one example with the interviewer before writing code; unclear prompts are a common complaint.
  • Keep C++ fresh (classes, inheritance, STL containers) if the role lists it, even when you expect Python.
  • Practise tree and interval problems to finish in about 30 minutes, leaving time for follow-ups.

A coding round in 13 ML, research and data science reports (six since 2024), mostly generic medium algorithm problems; one 2025 ML screen was C++ object-oriented coding.

04ML coding

About 45-60 min in Python, in the screen or the onsite

Implement ML building blocks or concepts in code. Reported prompts include implementing dropout, writing the forward and backward pass of softmax, and a 2025 ML-style problem that had a simpler trick the candidate found late. Research scientist onsites also list ML coding alongside standard coding.

What it tests

  • Implementing layers such as dropout correctly for training and inference
  • Deriving and coding gradients, for example softmax
  • Spotting a simpler formulation before writing a lot of code

How to prepare

  • Write dropout, softmax, cross-entropy and their backward passes in numpy from memory.
  • If your approach is getting long, pause and ask whether a simpler formulation exists.

ML coding described in 5 ML and research reports (three since 2024); examples given are dropout, a softmax backward pass and ML concepts in Python.

05ML and autonomy knowledge

About 60 min of discussion, in the screen (domain round) or the onsite as applied questions on Nuro's problems

A technical conversation on deep learning and the area you would work in. A 2024 research loop had two rounds of real-world questions from Nuro's field, such as how to tell when a scenario is uncertain for the driving agent; a data science screen covered high-level AI questions tied to the role. Older reports describe these rounds as jumping between topics.

What it tests

  • Deep learning fundamentals and trade-offs
  • Uncertainty estimation for prediction and planning
  • Applying ML to concrete autonomy scenarios
  • Explaining how your background maps to the team

How to prepare

  • Review perception, prediction and planning basics and how uncertainty is estimated and used downstream.
  • Answer in short, structured points and check the interviewer's intent before going deep.

Domain or applied ML rounds in 5 ML, research and data science reports (three since 2024); only one gives a concrete question.

06ML and ML infrastructure design

About 45 min, in the screen; one 2026 screen ran it back to back with coding

Few reports describe this round. A 2022 senior ML engineer designed an object-detection model pipeline; a 2026 ML engineer was told to expect ML design and instead got ML infrastructure: an online inference pipeline with model versioning, feature fetching, rollouts and monitoring.

What it tests

  • End-to-end perception pipeline: data, labels, model, evaluation
  • Model serving: routing, versioning, safe rollout
  • Feature computation at inference time and latency budgets
  • Monitoring model quality and data drift

How to prepare

  • Prepare both a modelling design (an object-detection pipeline) and a serving design; ask which one the round covers.
  • State latency and availability targets early and tie each component to how you would monitor it.

Free reported questions for this stage

Two ML engineer reports, from 2022 and 2026; one modelling design and one ML infrastructure design. Too few to know how common the round is.

07Past work deep dive and research talk

A screen round or part of other rounds on your projects; research scientists give a research talk at the onsite

Nuro interviewers spend a lot of time on your own work. ML engineers report a full screen round on a recent publication or a project discussion before coding; research scientist onsites include a talk, and in one 2024 loop the talk was centered on a theme set by the team (uncertainty). Expect questions on why you made each choice.

What it tests

  • Depth on a recent paper or project
  • Motivation and trade-offs behind your methods
  • Connecting your research to autonomy problems

How to prepare

  • Prepare one publication or project to discuss for a full round, including weaknesses and next steps.
  • For research roles, ask whether the talk has a set theme and adapt your material to it.

Past-work or research discussion in 8 ML and research reports, but only three since 2024; a research talk is described in one 2024 research loop.

08Behavioral and leadership chat

Onsite; one behavioral round, and in one 2019 loop a sit-down with a co-founder

Thinly reported. A 2024 research loop had one behavioral round among eight; a 2019 ML engineer onsite included a conversation with a co-founder about the company and its vision, plus a lunch interview.

What it tests

  • Why Nuro and its approach to autonomy
  • Teamwork and handling disagreement
  • Questions you ask about the company's direction

How to prepare

  • Read recent company news and job postings for your team so your questions are specific.
  • Have two stories on collaboration across teams ready.

A behavioral round in one 2024 research report and a co-founder conversation in one 2019 ML report; other reports do not describe it.

Questions and themes reported most often

  • Discuss one of your recent publications or projects in depth (8 reports, past work deep dive and research talk)
  • Generic medium algorithm problem unrelated to driving (6 reports, coding rounds (screen and onsite))
  • Why Nuro, and why autonomy (2 reports, recruiter call)
  • Implement dropout or the softmax backward pass by hand (2 reports, ml coding)
  • Object-oriented coding in C++ (2 reports, coding rounds (screen and onsite))
  • SQL queries on given tables, and SQL logic in Python (2 reports, online test (data science, some software roles))
  • Maximum path sum in an N-ary tree (1 reports, coding rounds (screen and onsite))
  • Design a production inference pipeline with versioning and monitoring (1 reports, ml and ml infrastructure design)

What Nuro asks about

  • General coding30%
  • Past work & research talk20%
  • ML coding15%
  • ML & autonomy concepts15%
  • ML / ML infra design10%
  • SQL (data science)5%
  • Behavioral5%

Nuro interview prep checklist

  • Solved tree, interval and dynamic programming problems at medium difficulty in about 30 minutes
  • Can write object-oriented C++ code as well as Python
  • Implemented dropout and a softmax backward pass from scratch
  • Prepared a deep dive on one publication or project, including weak points
  • Sketched both an object-detection pipeline and an online inference pipeline
  • Can explain how uncertainty is estimated and used in a driving stack
  • Asked the recruiter what each round covers and in which language

Tips

  • Treat the coding rounds as standard algorithm interviews; several candidates expected applied problems and did not get them.
  • Keep both Python and C++ ready, since one 2025 ML screen was in C++ and Nuro postings ask for both.
  • Write dropout, softmax and cross-entropy with their gradients in numpy until they are routine.
  • Prepare a project or paper you can defend for a full round; this comes up more than any single question.
  • Learn the basics of perception, prediction, planning and uncertainty estimation for driving.
  • Ask whether a design round is about modelling or about ML infrastructure and serving.
  • Clarify every prompt with an example before coding; unclear problem statements are a common complaint.
  • Expect slow or missing feedback; follow up after a week and keep other options open.

See the rest of the Nuro loop

The later stages, the questions reported for them and how to prepare are in Pro.

Unlock with Pro

Practice the Nuro loop

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

Nuro interview FAQ

How many interview rounds does Nuro have for ML roles?

Candidates report 8 stages: Recruiter call, Online test (data science, some software roles), Coding rounds (screen and onsite), ML coding, ML and autonomy knowledge, ML and ML infrastructure design, Past work deep dive and research talk and Behavioral and leadership chat. Most reports start with a recruiter call (about 30 minutes in one research report) and then two technical screens, sometimes back to back. The screens pair a coding round with a second round that varies: ML coding in Python, an ML or ML-infrastructure design, applied ML questions, or a talk about your research. Data science candidates in 2021 got an online or overnight SQL and Python test first. The onsite is described as six to eight one-on-one interviews: one or two coding rounds, ML coding, research or domain rounds tied to the team, a research talk for research scientists and a behavioral round; one 2019 onsite included a chat with a co-founder. One 2024 research candidate was told each interview's topic in advance. Several candidates say they never heard back after a round, so follow up and keep other processes moving.

How long does the Nuro ML interview process take?

Reported timelines run Not well reported; few candidates give a timeline.

What does Nuro ask in ML interviews?

By reported question mix: General coding 30%; Past work & research talk 20%; ML coding 15%; ML & autonomy concepts 15%; ML / ML infra design 10%; SQL (data science) 5%. The most repeated themes: Discuss one of your recent publications or projects in depth; Generic medium algorithm problem unrelated to driving; Why Nuro, and why autonomy.

How should I prepare for the Nuro ML interview?

Treat the coding rounds as standard algorithm interviews; several candidates expected applied problems and did not get them. Keep both Python and C++ ready, since one 2025 ML screen was in C++ and Nuro postings ask for both. Write dropout, softmax and cross-entropy with their gradients in numpy until they are routine.

Are there free Nuro interview questions?

Yes. LeetML has 2 questions reported from Nuro interviews, and 1 of them are free to solve in the browser; they are listed by stage above.

Related interview guides