Company interview guide
Waymo Machine Learning Interview Guide (2026)
Graph-heavy coding, numpy ML coding and driving-flavored ML design, plus a research talk for scientists.
Waymo's machine learning interview loop has 8 reported stages, from recruiter call to behavioral; typical length: 2+ weeks.
Roles covered: ML Engineer, Research Scientist and Deep Learning Engineer. The question bank holds 9 questions reported from Waymo interviews, 2 of them free to solve in your browser.
- Interview process length
- 2+ weeks
- Stages
- 8
- Reports linked to bank questions
- 9
- Bank report dates
- Nov 2025–Sep 2026
- Questions in bank
- 9 (2 free)
- Last reviewed
- October 2026
How the Waymo interview loop works
A recruiter call, one technical screen of about 45 minutes (usually data-structure coding; some research loops open with ML coding instead), then a virtual or same-day onsite of three to five interviews. The onsite usually mixes one or two coding rounds, ML coding, ML system design, ML fundamentals or modeling and a behavioral round; research loops add a research talk. Several ML engineer reports describe a two-week process, while research and UK loops have stretched to months with slow scheduling.
The Waymo interview process, stage by stage
01Recruiter call
Short call (about 30 min where stated); some candidates also had an informal coffee chat or a culture-fit check
A conversation about your background, the open roles that fit it and how the rest of the process will run. Research candidates in the UK also met the head of research before any technical round.
What it tests
- How your experience maps to an open team
- Why Waymo and why autonomous driving
- Logistics: location, work eligibility, level and timeline
How to prepare
- Ask which team the loop is for (perception, behavior prediction, planning, simulation, ML infrastructure); the design and modeling rounds follow the team.
- Have a short, specific answer to why Waymo; it was asked again inside technical rounds.
A recruiter call opens the process in 7 ML and research reports and 2 data scientist reports. Waymo's hiring page also describes a first call with a sourcer or recruiter.
02Technical screen
About 45 min, one problem live in a shared editor; research loops sometimes open with ML coding instead
Most reports describe one data-structures problem, often graph search (DFS, bidirectional BFS) or dynamic programming, at roughly medium difficulty. One 2026 research loop opened with an ML coding screen on self-attention, and older research loops ran two screens or a one-to-one technical interview.
What it tests
- Graph traversal on grids and implicit graphs
- Dynamic programming written cleanly and explained
- Turning an ambiguous prompt into a precise spec
- Randomized sampling tricks, such as building a uniform draw from coin flips
How to prepare
- Drill BFS and DFS variants, including bidirectional BFS and searches over sets of states.
- For DP, be ready to write the bottom-up table form; one interviewer insisted on it over memoized recursion.
- Ask clarifying questions before coding; a candidate who failed the screen called the questions simple but ambiguous.
A technical screen precedes the onsite in 11 ML and research reports; 8 describe data-structure coding. Waymo's hiring page also describes 1-2 phone or video interviews with a peer or manager.
03Onsite: coding
One or two data-structure rounds in the onsite, Python accepted
Graph problems with a twist dominate: BFS variants, a maze where a robot that cannot see must reach the exit with a single instruction string, and a build-once, query-many prefix lookup. Most rounds are reported as easy to medium, but an unfamiliar problem can sink an otherwise good loop. Two 2026 research loops had no data-structure round after the screen.
What it tests
- BFS and DFS over grids and over sets of possible states
- Choosing a data structure for repeated queries (for example a trie)
- Stating time and space complexity
- Working through a novel problem out loud
How to prepare
- Practise problems where the search state is a set of positions rather than one cell.
- If a problem is unfamiliar, start from a brute force and narrow it; one candidate wrote no code after waiting for an insight.
- Expect variations on classic interview problems rather than verbatim copies.
Onsite coding rounds appear in 11 ML and research reports, often graph or BFS-style problems. Waymo's hiring page also describes coding assessments and technical interviews in the onsite.
04ML coding
One round in numpy or PyTorch, sometimes the screen itself; one candidate received preparation material beforehand
Implement a model piece from scratch: softmax cross-entropy forward and backward plus a training loop, self-attention with a KV cache, patchifying images to train a small encoder, or sampling points inside a bounding box. A candidate for an ML-runtime team instead debugged a small numpy tensor utility with aliasing and axis bugs.
What it tests
- Forward and backward passes by hand, without autograd
- Vectorized numpy for image patches and box geometry
- Attention and incremental decoding with a cache
- Finding subtle bugs: shared row references, wrong stack axis, dropped remainders
How to prepare
- Write softmax cross-entropy and linear-layer gradients from scratch, then wrap them in a training loop.
- Practise reshape-based patchify and box sampling with no Python loops.
- Behavior prediction candidates: public write-ups (not candidate reports) mention vectorized minADE and minFDE over several trajectory hypotheses.
- If the recruiter sends preparation material, study it closely; the round may build on it.
Free reported questions for this stage
ML coding is described in 7 ML and research reports from 2025-2026, as the screen in one or two and an onsite round in the rest; tasks vary by team.
05ML system design
One open-ended discussion round; the setting follows the team
Usually an autonomy problem: design a robust vision system, a lane-following system, or a model that decides stop or go at a crosswalk from several sensors. An ML-systems candidate got inference serving for 100M daily users with back-of-the-envelope memory and bandwidth estimates. Several candidates found the prompts vague and the interviewer steering hard.
What it tests
- Framing inputs and outputs: cameras, lidar, radar, maps and what the model must emit
- Data and labels, especially for rare and safety-critical cases
- Offline and on-road evaluation, with metrics tied to safety
- Latency and compute budgets onboard or in serving
How to prepare
- Agree on scope and the expected depth in the first minutes; reports describe vague prompts and frequent redirects.
- Keep a template ready: sensors, representation, model, outputs, metrics, long-tail handling.
- Keep any recap of past work short; one candidate spent 15 minutes on it and was cut off.
Free reported questions for this stage
- Back-of-the-Envelope Design for Serving a Vision Model to 100M Daily UsersML System DesignHardSolve
An ML design round appears in 8 ML and research reports, often a driving, vision or serving prompt. Waymo's hiring page also lists design among its live interviews.
06ML fundamentals & modeling
One or two conversation rounds, sometimes built around your past projects
Labelled ML concepts, ML modeling, ML technical background or ML/DL theory depending on the loop. A 2026 ML engineer round was a project conversation with probing follow-ups; an ML-runtime loop asked about quantization-aware training, distillation, eval mode, contrastive learning and kernel fusion. Older research questions included zero-shot learning and the fastest way to convolve an image.
What it tests
- Deep learning fundamentals: losses, regularization, normalization, training behavior
- Vision basics for perception roles
- Model efficiency: quantization, distillation, kernels
- Defending the modeling choices in your own projects
How to prepare
- Review vision and deep learning fundamentals until you can explain each choice in your projects from first principles.
- ML-runtime and onboard roles: prepare compression, memory layout and kernel fusion trade-offs.
A fundamentals, theory or modeling round is named in 9 ML and research reports, consistently since 2025; content leans on deep learning and vision.
07Research talk & deep dive
Research scientist and research engineer loops: a talk or presentation, or a round about your own research
Research loops include a research talk, a presentation, or one or two interviews about your research; UK loops also included a meeting with the head of research. Engineer loops cover projects inside other rounds instead.
What it tests
- Clear motivation and contribution of your work
- Experimental rigor and honest limitations
- How your research would transfer to driving problems
How to prepare
- Prepare a talk a mixed audience can follow, and link it to perception, prediction or planning.
- Ask the recruiter for the talk length and who will attend.
A research talk, presentation or research interview appears in 5 research scientist and research engineer reports (2021-2026).
08Behavioral
One behavioral round in most recent onsites
Motivation, collaboration and how you work through ambiguity. Why Waymo also comes up inside technical rounds.
What it tests
- Why Waymo and why autonomous driving
- Collaboration and disagreement
- Working on ill-defined problems
How to prepare
- Connect your motivation to safety and to the team you are joining.
- Prepare two or three stories with concrete trade-offs and your own contribution.
A behavioral round is listed in 4 ML and research reports from 2025-2026. Waymo's hiring page also says behavioral and case-based questions run through the screens and the onsite.
Questions and themes reported most often
- Deep dive on a past project or your research, with probing follow-ups (6 reports, research talk & deep dive)
- Graph search problems: BFS, DFS, bidirectional BFS and mazes (5 reports, onsite: coding)
- Design a driving or vision system: robust perception, lane following, crosswalk stop-or-go (3 reports, ml system design)
- Implement attention or a ViT-style patch encoder from scratch (2 reports, ml coding)
- Vision geometry in numpy: bounding boxes and convolution (2 reports, ml coding)
- Why Waymo (2 reports, behavioral)
- Softmax cross-entropy backward pass, then a training loop (1 reports, ml coding)
- Uniform random draw built from a Bernoulli source (1 reports, technical screen)
What Waymo asks about
- Data structures & algorithms30%
- ML coding (numpy / PyTorch)20%
- ML system design (autonomy)20%
- ML & deep learning fundamentals15%
- Research & project deep dive8%
- Behavioral7%
Waymo interview prep checklist
- Solved several BFS and DFS variants, including one over sets of states, in under 45 minutes
- Wrote softmax cross-entropy forward and backward in numpy and trained a small model with it
- Implemented self-attention with a KV cache and a patchify step without loops
- Can sketch a perception or driving-decision system from sensors to safety metrics
- Know which team the loop is for and prepared that team's domain basics
- Prepared a project or research deep dive, including what did not work
- Have a specific answer to why Waymo
Tips
- Make graph search your first coding priority; BFS-style problems are the most often named question type.
- Ask early which team the loop is for; the design, modeling and ML coding rounds take that team's flavor.
- Get fluent in numpy without loops: patches, boxes, convolution and hand-written backward passes.
- For design, bring a driving template that ends in safety-relevant metrics and a plan for rare scenarios.
- Behavior prediction is unconfirmed by candidate reports: prepare multi-modal trajectory outputs and displacement metrics, which Waymo's public motion benchmarks center on.
- Clarify every ambiguous prompt; several candidates felt interviewers had one specific answer in mind.
- Research candidates: rehearse the talk and the follow-ups as seriously as the coding.
- Build slack into your plans; research and UK loops took weeks to months, and one candidate noted late interviewers.
See the rest of the Waymo loop
The later stages, the questions reported for them and how to prepare are in Pro.
Unlock with ProPractice the Waymo loop
Work the stages in order: every reported question for this company, grouped by round, with solutions in the app.
Waymo interview FAQ
How many interview rounds does Waymo have for ML roles?
Candidates report 8 stages: Recruiter call, Technical screen, Onsite: coding, ML coding, ML system design, ML fundamentals & modeling, Research talk & deep dive and Behavioral. A recruiter call, one technical screen of about 45 minutes (usually data-structure coding; some research loops open with ML coding instead), then a virtual or same-day onsite of three to five interviews. The onsite usually mixes one or two coding rounds, ML coding, ML system design, ML fundamentals or modeling and a behavioral round; research loops add a research talk. Several ML engineer reports describe a two-week process, while research and UK loops have stretched to months with slow scheduling.
How long does the Waymo ML interview process take?
Reported timelines run 2+ weeks (2 weeks to several months).
What does Waymo ask in ML interviews?
By reported question mix: Data structures & algorithms 30%; ML coding (numpy / PyTorch) 20%; ML system design (autonomy) 20%; ML & deep learning fundamentals 15%; Research & project deep dive 8%; Behavioral 7%. The most repeated themes: Deep dive on a past project or your research, with probing follow-ups; Graph search problems: BFS, DFS, bidirectional BFS and mazes; Design a driving or vision system: robust perception, lane following, crosswalk stop-or-go.
How should I prepare for the Waymo ML interview?
Make graph search your first coding priority; BFS-style problems are the most often named question type. Ask early which team the loop is for; the design, modeling and ML coding rounds take that team's flavor. Get fluent in numpy without loops: patches, boxes, convolution and hand-written backward passes.
Are there free Waymo interview questions?
Yes. LeetML has 9 questions reported from Waymo interviews, and 2 of them are free to solve in the browser; they are listed by stage above.
Related interview guides
- TeslaProject deep dives, practical numpy coding on driving data and layers written from scratch, with a loop shaped by the team you join.
- ZooxCoding with language and OOP questions, rapid-fire ML and vision fundamentals, and an unusual math and reasoning round.
- NuroAlgorithm coding, hand-written ML building blocks and a deep dive on your own research, for an autonomous driving company.
- Applied IntuitionCoding-heavy screens, hands-on transformer debugging and a fast loop, for an autonomy software company that is open about its pace.
- GoogleClassic algorithm rounds plus ML design, ML domain depth and Googleyness, judged by a hiring committee.