Company interview guide

Airbnb Machine Learning Interview Guide (2026)

Marketplace ranking design, a hard look at your past models, one practical coding round and a values check.

Airbnb's machine learning interview loop has 7 reported stages, from recruiter call to data science track: stats, experiments & take-home; typical length: 3-12 weeks.

Roles covered: ML Engineer and Data Scientist. The question bank holds 22 questions reported from Airbnb interviews, 2 of them free to solve in your browser.

Try a free question

Interview process length
3-12 weeks
Stages
7
Reports linked to bank questions
36
Bank report dates
Jan 2020–Sep 2026
Questions in bank
22 (2 free)
Last reviewed
October 2026

How the Airbnb interview loop works

A recruiter call, one live technical screen, then a virtual onsite of about four rounds. In all three recent ML engineer onsites that meant one coding round and two ML design rounds, plus either a project deep dive or a core values round. One recent senior candidate heard back two weeks after the onsite. Data scientist loops have used a take-home data challenge, a presentation, SQL, live data analysis, metrics and cross-functional values interviews. Several candidates describe slow scheduling, long silences and little feedback, with two data scientists reporting about three months end to end.

The Airbnb interview process, stage by stage

01Recruiter call

15-30 min phone call

A conversation about your background, the role and why Airbnb. Reported recruiter calls were short and scripted; one candidate was asked how they think about diversity, and another was asked why Airbnb repeatedly across the process.

What it tests

  • A genuine reason for wanting to work at Airbnb
  • How your past work maps to the role and track
  • Logistics: level, location, timeline

How to prepare

  • Ask which rounds the onsite contains and whether the track is ML engineering, data science or something in between.
  • Prepare a short, specific answer to why Airbnb and a story for each of the four core values; both come up more than once.

A recruiter or sourcer call opens the process in 8 data scientist reports (2017-2024); lengths given were 15 and 30 minutes. No ML engineer report describes this call.

02Technical screen

One live round of 30-60 min: ML engineers solve a coding problem with runnable tests; data scientists get SQL plus a metrics case or a Python data task

The one recent ML engineer screen was the layover problem: the fewest bookable experiences that exactly fill a given number of hours, which is unbounded coin change on decimal inputs (practise it under Onsite: coding, where it is reported twice more). The candidate finished in about 20 minutes, then printed the chosen combination as a follow-up. Data scientist screens mixed SQL with a metrics or product case, or ran a data-transformation task in an online editor; one 2020 loop added an online ML assessment first.

What it tests

  • Recognising a classic dynamic programming pattern inside a product story
  • Handling decimal inputs safely by scaling to integers
  • Reconstructing the chosen combination, not just the count
  • For data science: clean SQL plus a clear way to measure a business outcome

How to prepare

  • Never run DP directly on floats; multiply by ten or a hundred and work in integers.
  • Keep a parent array so you can print which items were used when the follow-up asks for it.
  • Run your own test cases before saying you are done; the recent ML screen expected passing tests.
  • For data science screens, practise three standard SQL questions plus a metric design question in 45 minutes.

One live screen in 9 reports, but only one is an ML engineer screen (2026, coding); the other 8 are data scientist screens from 2019-2024 with SQL, metrics or data-task content.

03Onsite: coding

1 live coding round in each recent ML engineer onsite

A single coding round in all three recent ML engineer onsites. Prompts are practical and specified tightly: the layover version of coin change in two of them (once with float denominations, where the interviewer flagged precision) and document redaction where every replacement is applied against the original text at once.

What it tests

  • Careful handling of string boundaries, overlaps and priority rules
  • Choosing a simple correct scan before reaching for clever structures
  • Spotting production details such as precision or re-processing of inserted text
  • Explaining time and space cost

How to prepare

  • Write the edge cases down before coding: overlapping matches, empty replacements, matches at the start and end, targets that cannot be formed.
  • Build output separately from the input so inserted text is never scanned again.
  • Scale decimal amounts to integers before any DP, and say why before the interviewer asks.

Exactly one coding round in all 3 ML engineer onsite reports from 2026; two used the coin change layover problem and one used phrase redaction.

04Onsite: ML experience deep dive

One discussion round on a project of your own; ML engineers often get it as the behavioral round

You walk through a recommendation or prediction project and the interviewer keeps asking why. Expect to defend model choice against simpler and more complex options, the decision threshold, latency and cost trade-offs, and how you scoped the work and resolved a disagreement.

What it tests

  • Why this model, and why not a more complex one
  • Trading off precision, recall, latency and infrastructure cost
  • What to do when offline metrics and online business metrics disagree
  • Threshold tuning and model comparisons such as factorization machines versus embedding networks
  • Scope, milestones and your personal contribution

How to prepare

  • Pick a project where you owned decisions end to end, and know the numbers behind each one.
  • Have a clear answer for threshold choice: candidates from the PR curve offline, final pick by an online test.
  • Separate the ML outcome from the product outcome, and do not invent conflict or results.

Two of three 2026 ML engineer onsites had a project deep dive that doubled as the behavioral round; one 2025 data scientist report describes project rounds ending in an open problem.

05Onsite: ML system design

2 open discussion rounds

The heaviest part of the loop: every recent ML engineer onsite had two of these rounds. Reported prompts are Airbnb problems: rank homes for a structured search with dates, guests and a map area but no text query; build a family-friendly listing filter when host labels are sparse and gamed; pick a cover photo or judge listing quality to lift bookings; and a RAG assistant that turns versioned support policies into steps for an agent.

What it tests

  • Two-stage retrieval and ranking with user and listing embeddings
  • Label design from bookings, reviews and complaints, and handling position bias in logs
  • Robustness to hosts who mislabel listings, plus fairness and cold start
  • Offline metrics, online experiments and marketplace guardrails
  • Grounded generation with policy versions, access rules and escalation

How to prepare

  • Start from the business goal (bookings, satisfaction, fewer bad stays) and define guardrails before the model.
  • Be ready for deep follow-up on position bias: randomised exploration traffic, inverse propensity weighting, stronger labels such as bookings.
  • Say which features exist at ranking time and which are future outcomes that would leak.
  • One senior candidate was turned down after a solid loop and felt hands-on production ranking experience was the deciding factor.

Free reported questions for this stage

Two ML design rounds in each of the 3 ML engineer onsite reports from 2026, all on marketplace problems; one 2025 data scientist report also lists a system design round.

06Core values interview

1-2 conversations, sometimes with people outside the hiring team

Airbnb has four values (champion the mission, be a host, embrace the adventure, be a cereal entrepreneur), and some loops test them in a dedicated round. One 2026 ML onsite ended with personal questions: why Airbnb, what you see as one of the biggest problems in the world, and which cause you care about and how you act on it. A 2019 data scientist onsite had two 30-minute cross-functional values chats. The other two recent ML onsites listed no separate values round, so ask your recruiter.

What it tests

  • A specific, personal connection to belonging and the mission
  • Concrete actions behind the values you claim
  • Care for colleagues and handling setbacks
  • Collaboration across functions

How to prepare

  • Read the core values on the careers site and tie one real story to each.
  • Answer the personal questions with things you actually do, not abstract beliefs.
  • Expect to be asked why Airbnb more than once; keep the answer consistent and specific.

A values round appears in one 2026 ML engineer onsite and one 2019 data scientist onsite; two other 2026 ML onsites had none. Public guides say it can block an offer.

07Data science track: stats, experiments & take-home

Older loops: take-home data challenge (72 hours in two reports) then onsite; recent loops: presentation, SQL, live data analysis and metrics rounds

Data scientist loops replace ML design with analysis work. Reports from 2017-2020 describe a take-home on messy, realistic data, submitted as a write-up and then presented onsite with a past project. The 2022 and 2024 reports instead describe a presentation of a past project or a short prepared deck, SQL, a live analysis of a dataset with the interviewer, and metrics or analytics cases. Public question lists add marketplace A/B testing, probability and regression, without saying which round they belong to.

What it tests

  • Why experiments are hard on a two-sided marketplace and how to design around it
  • Sample size, test duration and the risk of peeking at p-values
  • Estimating effects when an A/B test is impossible
  • Probability and estimation basics, including maximum likelihood
  • Turning an open dataset into a clear business recommendation

How to prepare

  • Treat any presentation as a memo for decision makers; one 2024 candidate kept the deck as short as instructed and was marked down for lack of depth.
  • If you get a take-home, budget a full weekend; candidates spent two to three days and one wrote an 8-page memo.
  • Practise metric design questions out loud, such as how to measure the customer support team.
  • Know interference between guests and hosts, long booking lead times and switchback or market-level tests.

Free reported questions for this stage

9 data scientist reports: 6 describe a take-home (2017-2020), 5 a presentation, 3 a live dataset analysis; the 2022 and 2024 loops mention no take-home.

Questions and themes reported most often

  • Take-home data challenge on messy booking data (data science, 2017-2020) (6 reports, data science track: stats, experiments & take-home)
  • Presentation of a past project or a prepared deck (data science) (5 reports, data science track: stats, experiments & take-home)
  • SQL plus metric design in the data science screen (4 reports, technical screen)
  • Fewest experiences or coins that exactly fill a layover or amount, with decimal inputs (3 reports, onsite: coding)
  • Live analysis of a dataset with the interviewer (data science) (3 reports, data science track: stats, experiments & take-home)
  • Why Airbnb, the biggest problem in the world, and your personal cause (3 reports, core values interview)
  • Rank homes for a structured search with no text query, two-stage retrieval and ranking (2 reports, onsite: ml system design)
  • Defend a past model: complexity, thresholds, offline versus online gaps (2 reports, onsite: ml experience deep dive)

What Airbnb asks about

  • ML system design35%
  • Coding20%
  • ML experience & judgment15%
  • Stats & experimentation15%
  • Core values & behavioral15%

Airbnb interview prep checklist

  • Solved the layover coin change on decimal inputs and printed the combination in under 20 minutes
  • Can sketch a two-stage home search ranker and explain three ways to correct position bias
  • Designed a listing classifier that copes with sparse, gamed host labels and cold start
  • Can defend one past model on complexity, threshold, cost and offline versus online gaps
  • Have a real story for each of the four core values and a specific answer to why Airbnb
  • For data science: can present a past project in 15 minutes and explain why marketplace A/B tests are hard

Tips

  • Make two-stage home search ranking your first design to master, including position bias and marketplace guardrails.
  • Prepare one recommendation or prediction project to defend in depth: model choice, threshold, cost, and an offline versus online disagreement.
  • Drill unbounded coin change until you can handle decimals and print the chosen items without thinking.
  • Write runnable code and your own test cases in every coding practice session.
  • Frame every design around bookings and guest and host experience, not just click-through rate.
  • Prepare the core values round as seriously as a technical one, even though not every loop includes it.
  • For data science roles, practise a short presentation of a past project and a live analysis of an unfamiliar dataset, plus marketplace A/B testing questions.
  • Expect slow scheduling and sparse feedback, and keep other processes moving in parallel.

See the rest of the Airbnb loop

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

Unlock with Pro

Practice the Airbnb loop

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

Airbnb interview FAQ

How many interview rounds does Airbnb have for ML roles?

Candidates report 7 stages: Recruiter call, Technical screen, Onsite: coding, Onsite: ML experience deep dive, Onsite: ML system design, Core values interview and Data science track: stats, experiments & take-home. A recruiter call, one live technical screen, then a virtual onsite of about four rounds. In all three recent ML engineer onsites that meant one coding round and two ML design rounds, plus either a project deep dive or a core values round. One recent senior candidate heard back two weeks after the onsite. Data scientist loops have used a take-home data challenge, a presentation, SQL, live data analysis, metrics and cross-functional values interviews. Several candidates describe slow scheduling, long silences and little feedback, with two data scientists reporting about three months end to end.

How long does the Airbnb ML interview process take?

Reported timelines run 3-12 weeks.

What does Airbnb ask in ML interviews?

By reported question mix: ML system design 35%; Coding 20%; ML experience & judgment 15%; Stats & experimentation 15%; Core values & behavioral 15%. The most repeated themes: Take-home data challenge on messy booking data (data science, 2017-2020); Presentation of a past project or a prepared deck (data science); SQL plus metric design in the data science screen.

How should I prepare for the Airbnb ML interview?

Make two-stage home search ranking your first design to master, including position bias and marketplace guardrails. Prepare one recommendation or prediction project to defend in depth: model choice, threshold, cost, and an offline versus online disagreement. Drill unbounded coin change until you can handle decimals and print the chosen items without thinking.

Are there free Airbnb interview questions?

Yes. LeetML has 22 questions reported from Airbnb interviews, and 2 of them are free to solve in the browser; they are listed by stage above.

Related interview guides

Official sources