Company interview guide

Nextdoor Machine Learning Interview Guide (2026)

Two back-to-back technical screens: practical coding plus a hands-on tabular ML pipeline, then a final stage of varying length.

Nextdoor's machine learning interview loop has 6 reported stages (Recruiter call, Coding round, ML coding: end-to-end pipeline, ML fundamentals and recommender design, Final round, Data scientist track: SQL, experiments and product sense); typical length: 2-4 weeks.

Roles covered: ML Engineer and Data Scientist. The question bank holds 4 questions reported from Nextdoor interviews.

Try a free question from this loop's topics

Interview process length
2-4 weeks
Stages
6
Reports linked to bank questions
6
Bank report dates
Jul 2025–Apr 2026
Questions in bank
4 (0 free)
Last reviewed
October 2026

How the Nextdoor interview loop works

A 30-minute recruiter call, then a technical screen of two 60-minute rounds, coding then ML, with a break of 30 to 60 minutes between them. Most rejections in our reports came at this screen. Candidates who pass describe a final stage that varies a lot: one 2026 report had a single 1.5-hour session of more coding plus behavioral questions, while 2023-2024 reports describe three virtual onsite rounds or about six hours of coding, ML coding and ML design. One candidate gave the whole process as about two weeks. Data scientist candidates instead report a SQL round followed by about four onsite conversations with the hiring manager, a stakeholder and a senior data scientist, focused on experiments and product success metrics. The company's careers pages do not describe the interview process.

The Nextdoor interview process, stage by stage

01Recruiter call

About 30 min by video with a recruiter; one report adds a separate hiring-manager call

The recruiter walks through the process, checks interest and logistics such as start date, and reviews your resume. Two candidates say it is more than a chat: expect real behavioral questions such as why Nextdoor, the project you are most proud of, a recent project, and even the last research paper you read.

What it tests

  • Why Nextdoor and why this team
  • A clear story about the project you are most proud of
  • Fit between your experience and product-facing ML
  • Logistics: start date, location, level

How to prepare

  • Prepare two or three resume projects with problem, your role, metric and outcome; the recruiter may probe them.
  • Have an answer ready for the last paper you read and what you took from it.
  • Ask which ML round you will get (pipeline coding or concepts) and which libraries are allowed.

A recruiter call opens the process in 8 ML-engineer reports from 2023 to 2026, usually about 30 minutes; in two of them the recruiter also asked behavioral questions.

02Coding round

60 min live coding, first round of the technical screen; some finals add another coding session

A medium-level problem that is usually a twist on a classic one: sorting version strings with a custom comparator, merging meeting intervals given as day-and-time strings, building a ranked feed that groups photos in threes, or turning parent-child comment pairs into a tree. Several candidates note the work is in the parsing and edge cases rather than a clever algorithm, and one called it a non-standard hashmap and recursion problem.

What it tests

  • Parsing messy string input into clean structures
  • Custom ordering and stable sorting
  • Tree building and BFS / DFS traversals
  • Interval logic and hash maps
  • Clean, working code within the hour

How to prepare

  • Practise the classic interval, comparator and tree problems, then add a parsing layer on top (day names, times, version tags).
  • State tie-breaking and edge-case rules out loud before you code; these problems hide their difficulty there.
  • Leave time to run through an example by hand; candidates who solved it slowly still failed.

A general coding round appears in 12 ML-engineer reports from 2022 to 2026, usually one 60-minute problem in the screen; two give tree problems and three give parsing-heavy variants.

03ML coding: end-to-end pipeline

60 min live, second round of the screen; one 2026 report allowed only pandas and numpy plus web docs, no AI tools

You get a tabular dataset and build the whole pipeline by hand: load the data, handle missing values, encode categoricals, split, train a classifier or regressor, and evaluate it. Library rules vary; one candidate had to write the model with pandas and numpy only, another wrote a training loop and forgot the optimizer step. Expect a stream of ML fundamentals questions while you code.

What it tests

  • Data cleaning: imputation, encoding, scaling, stratified splits
  • Implementing a simple model and training loop from scratch
  • Choosing and computing evaluation metrics
  • Explaining ML fundamentals while coding

How to prepare

  • Write a full tabular pipeline from memory with numpy only, including logistic regression trained by gradient descent.
  • If you use a framework, check the basics: zero gradients, call the optimizer step, switch to eval mode for scoring.
  • Avoid leakage: fit imputers and scalers on the training split only, and say so.

Hands-on ML coding in 5 ML-engineer reports from 2024 to 2026; three of them describe the same build-a-pipeline-from-a-dataset task in the screen.

04ML fundamentals and recommender design

About 60 min of discussion; the second screen round in reports up to early 2025, and a design round in later stages

Broad questions on core ML plus an open design problem, most often a recommender system. Reported prompts include designing a news feed recommender, an end-to-end recommendation pipeline and a fraud detection system, with concept questions on PR-AUC, bagging, overfitting and model evaluation. One 2026 candidate felt this team rewards crisp recall of fundamentals more than years of experience.

What it tests

  • Recommender pipeline: candidate generation, ranking, evaluation
  • Metrics for imbalanced problems such as PR-AUC
  • Ensembles, overfitting and the bias-variance trade-off
  • Trade-offs between modelling approaches for a product problem

How to prepare

  • Prepare one feed or notification recommender design end to end, with offline and online metrics and a feedback-loop discussion.
  • Drill short, precise definitions: PR-AUC vs ROC-AUC, bagging vs boosting, regularization, cross-validation.
  • Tie designs to a neighborhood social product: local content, sparse engagement, and not over-notifying users.

ML concept or recommender-design discussion in 8 ML-engineer reports from 2022 to 2026; recommender systems come up in four, and it left the screen for later rounds by mid-2025.

05Final round

Varies: 1.5 hours of coding plus behavioral in one 2026 report; three virtual rounds or about six hours of coding, ML coding and ML design in 2023-2024 reports

The final stage is the least consistent part of the loop. The latest report describes one short session mixing another coding problem with behavioral questions, while older reports describe a full virtual onsite with separate coding, ML coding and ML design rounds, and one included a hiring-manager call. Expect a project you are proud of to come up again.

What it tests

  • A second coding problem at similar difficulty
  • ML design depth on a feed or ranking problem
  • The project you are most proud of, in detail
  • Collaboration with product and data science

How to prepare

  • Ask the recruiter for the exact rounds; the final has ranged from 1.5 hours to a full day.
  • Prepare one project deep dive with the metric you moved and an experiment that went wrong.
  • Show you can own a model from data to live A/B test; the team is small and expects end-to-end work.

Only 3 ML-engineer reports describe the stage after the screen (2023, 2024, 2026), and they disagree on length; most candidates in our reports stopped at the screen.

06Data scientist track: SQL, experiments and product sense

A SQL round, then about four onsite conversations with the hiring manager, a stakeholder and a senior data scientist

Data scientist candidates report no Python coding round. After a SQL screen, the onsite covers experiment design, how you would measure a product's success and leadership questions, and candidates describe the questions as fair and fairly easy.

What it tests

  • SQL on product event data
  • A/B test design and interpretation
  • Defining success metrics for a product feature
  • Leadership and stakeholder communication

How to prepare

  • Practise turning a vague feature into a primary metric, guardrails and an experiment plan.
  • Expect a non-technical stakeholder in the loop; explain results in plain language.

Two data-scientist reports (2024 and 2025) describe this track: SQL, then hiring manager, stakeholder and senior data scientist rounds on experiments and product success.

Questions and themes reported most often

  • Design a recommender system, often for a news feed (4 reports, ml fundamentals and recommender design)
  • The project you are most proud of (4 reports, final round)
  • Build a tabular classification pipeline from a raw dataset: clean, encode, train, evaluate (3 reports, ml coding: end-to-end pipeline)
  • ML fundamentals: PR-AUC, bagging, overfitting and evaluation (3 reports, ml fundamentals and recommender design)
  • Tree problems: build a comment tree from parent-child pairs, BFS and DFS traversals (2 reports, coding round)
  • Why Nextdoor (2 reports, recruiter call)
  • How you would measure a product's success (2 reports, data scientist track: sql, experiments and product sense)
  • Sort version strings with a custom comparator (1 reports, coding round)

What Nextdoor asks about

  • General coding35%
  • ML coding (tabular pipeline)22%
  • ML fundamentals & recommender design23%
  • Behavioral & project deep dive12%
  • Product metrics & experiments (DS)8%

Nextdoor interview prep checklist

  • Built a CSV-to-metrics classification pipeline with numpy only, with no leakage
  • Solved interval merging, version sorting and a feed-grouping problem with full input parsing
  • Built a tree from parent-child pairs and traversed it both iteratively and recursively
  • Can explain PR-AUC vs ROC-AUC, bagging vs boosting and how to spot overfitting in one minute each
  • Sketched a news feed or notification recommender with offline and online metrics
  • Have a two-minute story for the project you are most proud of and a crisp why Nextdoor
  • Asked the recruiter about allowed libraries and the format of the final round

Tips

  • Treat the two-round screen as the main event: it is where most candidates in our reports were rejected.
  • Write a full tabular ML pipeline with numpy and pandas only, end to end, under 45 minutes.
  • Practise classic coding problems wrapped in string parsing: times with day names, version tags, typed feed items.
  • Know your fundamentals cold; candidates got many quick definition questions during the ML round.
  • Prepare one recommender design for a feed or notifications, with metrics and how you would A/B test it.
  • Ask the recruiter which tools are allowed in the ML round; one 2026 candidate could search docs but not use AI tools.
  • Treat the recruiter call as a real behavioral interview: why Nextdoor, proudest project, a recent paper.
  • Plan for a gap of 30 to 60 minutes between the two screen rounds and keep your energy for the second.

Practice the Nextdoor loop

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

Nextdoor interview FAQ

How many interview rounds does Nextdoor have for ML roles?

Candidates report 6 stages: Recruiter call, Coding round, ML coding: end-to-end pipeline, ML fundamentals and recommender design, Final round and Data scientist track: SQL, experiments and product sense. A 30-minute recruiter call, then a technical screen of two 60-minute rounds, coding then ML, with a break of 30 to 60 minutes between them. Most rejections in our reports came at this screen. Candidates who pass describe a final stage that varies a lot: one 2026 report had a single 1.5-hour session of more coding plus behavioral questions, while 2023-2024 reports describe three virtual onsite rounds or about six hours of coding, ML coding and ML design. One candidate gave the whole process as about two weeks. Data scientist candidates instead report a SQL round followed by about four onsite conversations with the hiring manager, a stakeholder and a senior data scientist, focused on experiments and product success metrics. The company's careers pages do not describe the interview process.

How long does the Nextdoor ML interview process take?

Reported timelines run 2-4 weeks.

What does Nextdoor ask in ML interviews?

By reported question mix: General coding 35%; ML coding (tabular pipeline) 22%; ML fundamentals & recommender design 23%; Behavioral & project deep dive 12%; Product metrics & experiments (DS) 8%. The most repeated themes: Design a recommender system, often for a news feed; The project you are most proud of; Build a tabular classification pipeline from a raw dataset: clean, encode, train, evaluate.

How should I prepare for the Nextdoor ML interview?

Treat the two-round screen as the main event: it is where most candidates in our reports were rejected. Write a full tabular ML pipeline with numpy and pandas only, end to end, under 45 minutes. Practise classic coding problems wrapped in string parsing: times with day names, version tags, typed feed items.

Are there free Nextdoor interview questions?

LeetML has 4 questions reported from Nextdoor 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