Wayve interview, decoded.

Wayve prints its interview loop in its job ads: a 30-minute recruiter screen, two hours of competency interviews on programming and system design, an hour of domain deep dive, and a final hour on the mission and values. For machine learning engineers the round people talk about is the PyTorch one: candidates on Glassdoor describe an hour with a Google Colab notebook that trains a convolutional network, where the job is to find and fix what is broken. Software engineers report a live coding round with a technical lead and a broader architecture discussion.

Wayve trains cars to drive from data rather than hand-written rules. It raised a $1.2B Series D at an $8.6B valuation in February 2026, led by Eclipse, Balderton and SoftBank with Microsoft, NVIDIA, Uber and three carmakers taking part, and its robotaxis went live on Uber in London in September 2026 with a safety driver on board.

Free · 2 minutes · no account

The questions for your exact Wayve role and level.

See the questions Wayve asks ↓Run a free mock interview →

Interviewing at Wayve? Below are the questions candidates report. For the ones your exact role and level will get, paste the posting. Your first mock is free.

01

Who Wayve hires

Different tracks, one bar

Wayve hires machine learning engineers and research scientists (perception, vision models, reinforcement learning, foundation models for driving), software engineers for the data platform, fleet management, evaluation tools and the software that runs on the car, embedded and systems engineers, and site reliability engineers. The largest engineering groups are in Sunnyvale and London, with smaller teams in Leonberg, Tokyo, Herzliya, Detroit and Vancouver. Offices are hybrid, with time in workshops and labs alongside the vehicles.

The ads give one loop for every engineering role. What the technical hours contain depends on which side of the company you join.

  • Track · ML

    Machine learning engineering

    A programming round on data structures and algorithms, the PyTorch debugging hour in a Colab notebook, and a deep learning case study, per Glassdoor. The domain deep dive follows your team: perception, 3D and tracking for ADAS, vision models, reinforcement learning.

  • Track · SWE

    Software engineering

    A live coding round of about an hour with a technical lead, then an architecture discussion on scale and open-ended product problems. Data roles work at the scale of 500,000 hours of driving data, hundreds of petabytes, so expect the design questions to live there.

02

What is the Wayve interview process?

Round by round

The four stages are Wayve's own, from its machine learning engineer ads published in July 2026. The round contents come from Glassdoor reviews and candidate accounts. Glassdoor puts the average at about 24 days; one account puts it at two weeks.

01
Recruiter screen
30 minutes

Your background, salary expectations, and why Wayve's approach to driving interests you.

02
Competency interviews
Two hours: programming and system design

A live coding round on data structures and algorithms, and a system design round. Software engineers report the coding led by a technical lead and judged on code quality and reasoning, and the design round covering architecture, scale and open-ended product problems.

03
PyTorch debugging
About an hour, in a Google Colab notebook

A notebook that trains a CNN and does not work. Find the errors, fix them, and explain what each one did to training. The section below goes through what to check.

Machine learning roles
04
Domain deep dive
One hour

Your specialism against the team's problems. ML candidates report a deep learning case study; one reported question asks how you would find every example in a dataset of telemetry and video where the car had to brake hard because another vehicle cut in.

05
Mission and values
One hour, often with a director of engineering

Why embodied AI, how you work, and whether you want to help build a company whose playbook, in its own ads, is still being written.

03

What Wayve screens for

What every round is really testing

Wayve's ads close with a warning that it is not a settled company, and the final interview checks you are fine with that.

  • Ownership where the playbook is still being written
  • Pragmatic, measurable progress on real driving performance
  • End-to-end work across data, training, evaluation and iteration
  • Care for safety on real roads
  • Mission alignment with embodied AI
04

Wayve interview questions

Candidate-reported themes

Candidates describe the formats in more detail than the questions. The reported ones are below; the technical themes come from the reports and the ads.

Behavioural & motivation

  • Why Wayve, and why embodied AI?
  • Tell me about a model or system you took from data to production.
  • How do you decide what to fix next when a model underperforms?
  • Tell me about working somewhere without a settled process.

Technical

  • PyTorch debugging: A broken CNN training notebook in Colab: find the faults, fix them, explain them
  • Data mining for driving: Reported: find every emergency brake caused by a cut-in across telemetry and video
  • Perception: Detection, segmentation, 3D and tracking, and auto-labelling pipelines for ADAS
  • Data platform design: Ingesting hundreds of petabytes of driving data, where one bad segment must not block the pipeline
  • Data structures and algorithms: A live coding hour in the competency interviews

These are the reported themes — your loop is role-specific. Paste the actual posting and Calibrd predicts the questions for that exact role and level.

Predict my Wayve questions →
05

Wayve PyTorch debugging interview: what to check in the notebook

A checklist for the round

Machine learning candidates describe one hour with a Google Colab notebook that trains a convolutional network and has errors to find and fix. Nobody has published the bugs themselves, and they will change. These are the faults a broken training loop usually contains, in the order worth checking, so you have a routine instead of a guess.

  • Does it run at all?: Read the first stack trace carefully. Shape mismatches between the last convolution and the first linear layer, and tensors on different devices (one on the GPU, one on the CPU), are the usual crashes.
  • Is the data right?: Show a few images with their labels before touching the model. Look for labels that do not match, a missing or doubled normalisation, channels in the wrong order, and a train set that is not shuffled.
  • Is the loss right for the output?: Cross-entropy in PyTorch expects raw scores and class indices; a softmax before it, or one-hot targets, quietly breaks training. Check the loss is averaged the way you think.
  • Is the training step in the right order?: zero_grad, forward, loss, backward, step. A missing zero_grad lets gradients pile up; a step before backward does nothing useful.
  • Are train and eval modes switched?: model.train() for training and model.eval() with torch.no_grad() for evaluation, or dropout and batch norm behave wrongly and the validation numbers lie.
  • Can it overfit one batch?: Train on a single small batch. If the loss will not go near zero, the bug is in the model or the loop, not the data size. Then check the learning rate and the optimizer's parameter list.

Say what each bug did to training as you fix it. The round is as much about how you reason through a broken run as about the fixes.

06

Pay

What the offer looks like
Software Engineer, Greater London£129K median total pay
US ML engineer, senior (posted)$311,850–$389,400 base plus equity
US software engineer, senior (posted)$209,700–$266,800 base plus equity

Levels.fyi puts the median software engineer package at Wayve in Greater London at £129K, updated in August 2026 and read through search results on 23 September 2026. Wayve's UK ads mostly carry no range; its Sunnyvale ads do, from $176,500 for a fleet management software engineer to $311,850 to $389,400 for a senior machine learning engineer, each with equity. The ads also promise salaries benchmarked yearly, relocation and visa sponsorship where it applies. Calibrd benchmarks the pay on any posting you scan against its level and location.

07

How to prepare for a Wayve interview

In order
  1. 01Break a PyTorch training loop on purpose, then fix it against a clock. Run through data, loss, step order, modes and overfitting one batch until it is a habit.
  2. 02Revise data structures and algorithms for a live hour. The competency interviews include one for every engineering role.
  3. 03Prepare a design for driving data at scale: ingesting petabytes, letting one corrupt segment fail without blocking the rest, and making it searchable for rare events.
  4. 04Practise mining a dataset for rare events, like the reported emergency brake question: what signals you would use, how you would check the results.
  5. 05Read about end-to-end driving before the values round and have a view on it. The final hour is about the mission.

This guide covers Wayve's machine learning and software engineering loops. Hardware, fleet operations and field roles run a different process. For management and leadership roles the loop is similar but the bar shifts to people, delivery and strategy, so pair it with the leadership interview prep hub. The bar for your exact role comes from the role-by-role guides, and the prep that actually transfers is spoken, so run a mock interview before the real one.

Knowing the questions isn’t the same as answering them out loud. Run a Wayve mock: spoken answers, coached on the spot. Your first mock is free.

Run a Wayve mock →
08

FAQ & sources

The short answers
What is Wayve's interview process?

The four stages are Wayve's own, from its machine learning engineer ads published in July 2026. The round contents come from Glassdoor reviews and candidate accounts. Glassdoor puts the average at about 24 days; one account puts it at two weeks. Recruiter screen: Your background, salary expectations, and why Wayve's approach to driving interests you. Competency interviews: A live coding round on data structures and algorithms, and a system design round. Software engineers report the coding led by a technical lead and judged on code quality and reasoning, and the design round covering architecture, scale and open-ended product problems. PyTorch debugging: A notebook that trains a CNN and does not work. Find the errors, fix them, and explain what each one did to training. The section below goes through what to check. Domain deep dive: Your specialism against the team's problems. ML candidates report a deep learning case study; one reported question asks how you would find every example in a dataset of telemetry and video where the car had to brake hard because another vehicle cut in. Mission and values: Why embodied AI, how you work, and whether you want to help build a company whose playbook, in its own ads, is still being written.

What does Wayve look for in candidates?

Wayve's ads close with a warning that it is not a settled company, and the final interview checks you are fine with that. Ownership where the playbook is still being written Pragmatic, measurable progress on real driving performance End-to-end work across data, training, evaluation and iteration Care for safety on real roads Mission alignment with embodied AI

What questions does Wayve ask in interviews?

Candidates describe the formats in more detail than the questions. The reported ones are below; the technical themes come from the reports and the ads. Why Wayve, and why embodied AI? Tell me about a model or system you took from data to production. How do you decide what to fix next when a model underperforms? Tell me about working somewhere without a settled process. PyTorch debugging Data mining for driving Perception Data platform design Data structures and algorithms

What is the Wayve PyTorch debugging interview?

Machine learning candidates describe one hour with a Google Colab notebook that trains a convolutional network and has errors to find and fix. Nobody has published the bugs themselves, and they will change. These are the faults a broken training loop usually contains, in the order worth checking, so you have a routine instead of a guess. Does it run at all?: Read the first stack trace carefully. Shape mismatches between the last convolution and the first linear layer, and tensors on different devices (one on the GPU, one on the CPU), are the usual crashes. Is the data right?: Show a few images with their labels before touching the model. Look for labels that do not match, a missing or doubled normalisation, channels in the wrong order, and a train set that is not shuffled. Is the loss right for the output?: Cross-entropy in PyTorch expects raw scores and class indices; a softmax before it, or one-hot targets, quietly breaks training. Check the loss is averaged the way you think. Is the training step in the right order?: zero_grad, forward, loss, backward, step. A missing zero_grad lets gradients pile up; a step before backward does nothing useful. Are train and eval modes switched?: model.train() for training and model.eval() with torch.no_grad() for evaluation, or dropout and batch norm behave wrongly and the validation numbers lie. Can it overfit one batch?: Train on a single small batch. If the loss will not go near zero, the bug is in the model or the loop, not the data size. Then check the learning rate and the optimizer's parameter list.

How do I prepare for a Wayve interview?

Break a PyTorch training loop on purpose, then fix it against a clock. Run through data, loss, step order, modes and overfitting one batch until it is a habit. Revise data structures and algorithms for a live hour. The competency interviews include one for every engineering role. Prepare a design for driving data at scale: ingesting petabytes, letting one corrupt segment fail without blocking the rest, and making it searchable for rare events. Practise mining a dataset for rare events, like the reported emergency brake question: what signals you would use, how you would check the results. Read about end-to-end driving before the values round and have a view on it. The final hour is about the mission.

Prep for a real Wayve role

Practise your Wayve interview, out loud.

Paste a real Wayve posting and Calibrd predicts the questions for that role and level, benchmarks the pay, and flags the gaps an interviewer will probe in your CV — then listens to your spoken answers and coaches them. Your first mock is free.

Free to start · No card · Encrypted at rest, never used to train AI, remove anytime

Two weeks out?

Email me the Wayve prep checklist

The round that decides it, the questions they ask, and what to do before, in one email. Nothing else unless you sign up.

Wayve Interview: PyTorch Debugging Round, ML Loop — Calibrd