Company interview guide

Zillow Machine Learning Interview Guide (2026)

Real-estate flavoured ML and data loops: a hiring-manager or coding screen, then a mixed loop of coding, ML depth and cases.

Zillow's machine learning interview loop has 7 reported stages, from recruiter or hr call to behavioral & project deep dive; typical length: 2-8 weeks.

Roles covered: ML Engineer, Applied Scientist and Data Scientist. The question bank holds 13 questions reported from Zillow interviews, 5 of them free to solve in your browser.

Try a free question

Interview process length
2-8 weeks
Stages
7
Reports linked to bank questions
13
Bank report dates
Apr 2018–Nov 2025
Questions in bank
13 (5 free)
Last reviewed
October 2026

How the Zillow interview loop works

A recruiter or HR call, then one technical screen of about an hour, often with the hiring manager. Before about 2023 most data and applied science candidates also did a take-home modeling project, usually house-price prediction, before or after that screen; no report since 2023 mentions one. The final loop is about four or five interviews of 30-60 minutes, now held remotely: one 2024 ML engineer loop was three coding rounds and one behavioral round, while data science loops mixed SQL or R coding, business cases and behavioral rounds with managers from several teams. Applied scientist reports since 2024 mostly stop at the hiring-manager screen, so their final loop is thinly documented. Several candidates describe slow scheduling and little feedback after a rejection, so follow up on your own.

The Zillow interview process, stage by stage

01Recruiter or HR call

20-30 min phone or video call; recent applied scientist candidates were booked through an automated scheduling text

A background and logistics call: your experience, why Zillow, location or relocation, visa status and salary range. Some recruiters also ask a few basic ML questions, and one used the call to match the candidate to a team. Candidates often get little information about what the next rounds cover, so ask directly.

What it tests

  • A clear summary of your background and the role you want
  • Interest in Zillow and its housing business
  • Logistics: location, relocation, visa, salary range

How to prepare

  • Ask which skills the next round will test (SQL, coding language, ML theory) and what tooling it uses.
  • Have a short answer ready on why Zillow and how your work relates to housing or marketplace data.
  • Expect gaps in scheduling; send a polite follow-up if a week passes without news.

Recruiter or HR call first in 18 ML, data and applied science reports (4 since 2024) plus one 2024 engineering report; content is consistently background and logistics.

02Take-home modeling project (older loops)

Dataset and questions by email, usually about one week to submit code and a short report; one 2018 ML engineer got a timed online coding test instead

For years the signature Zillow data science step: predict house sale prices, Zestimate style, or analyse a real-estate dataset, then discuss it with the team. One candidate was rejected because the hiring manager wanted more figures in the report, so presentation matters as much as accuracy. No report since 2023 mentions a take-home, so check whether your team still uses one.

What it tests

  • Building and validating a price model on tabular housing data
  • Feature engineering, outlier handling and choice of error metric
  • Communicating results in a clear written report with figures
  • Explaining your choices when the screen asks about the project

How to prepare

  • Treat the report as the deliverable: state the question, show plots, compare models and explain the chosen one.
  • Use a validation scheme that respects location and time, and report an error metric a business reader understands.
  • Ask how long you are expected to spend and stay close to it; one candidate was told 5-10 hours.

Free reported questions for this stage

Take-home project or assignment in 19 data, applied science and ML reports from 2014 to 2022, most of them house-price prediction, plus one timed online test; none since 2023.

03Technical or hiring-manager screen

About 60 min by phone or video, with an engineer, a senior data scientist or the hiring manager; content depends on the track

ML engineers usually get one or two easy-to-medium data-structure problems in a shared editor (a bracket-validity variant, a heap-based streaming question), often with resume questions and, in 2025, LLM questions such as why models hallucinate and how to chunk documents for retrieval. Applied scientists mostly meet the hiring manager for ML and deep-learning theory, sometimes with a small design question. Data scientists get SQL on user and saved-home tables, sometimes with a product question about a falling metric. Interviewers may push for a different approach after you finish, so be ready with alternatives.

What it tests

  • Correct, tested code for an easy-to-medium algorithm problem
  • SQL joins, anti-joins and date arithmetic on product tables
  • ML fundamentals: regression, overfitting, regularization, CNNs and transformers
  • LLM application basics: hallucination causes and retrieval chunking
  • Depth on a project from your resume

How to prepare

  • Ask clarifying questions early and restate the problem; several candidates found prompts underspecified.
  • After a working solution, offer a second approach and its trade-offs before you are asked.
  • For SQL, use standard syntax rather than a vendor dialect your interviewer may not know.
  • For applied science screens, be able to explain linear regression, overfitting and attention from first principles.

Free reported questions for this stage

One technical or hiring-manager screen in 30 ML, data and applied science reports, 7 since 2024: coding for ML engineers, ML theory for applied scientists, SQL for data scientists.

04Final loop: coding

1-3 x about 60 min, remote; algorithm or practical coding for ML engineers, SQL and R or Python for data scientists, pseudocode accepted in some LLM rounds

In the one detailed 2024 ML engineer final, three of four rounds were coding. A 2025 ML engineer round asked for a ReAct-style agent loop in pseudocode rather than an algorithm puzzle. Data science finals used SQL plus a general coding question in R, such as checking whether two rectangles overlap. The company's own preparation page, written for software engineers, describes greenfield coding that mimics real work and a refactoring exercise rather than puzzle questions.

What it tests

  • Building a small working feature from a blank page
  • Clean structure, naming and edge-case handling
  • Writing an LLM tool-use loop with stopping rules
  • SQL and basic programming for data scientist tracks

How to prepare

  • Practise building small features end to end with tests, not only algorithm drills.
  • For agent questions, state the loop, tool dispatch, step budget and stop condition before writing code.
  • Watch the company's greenfield and refactoring videos; they describe what interviewers look for.

Free reported questions for this stage

Coding in the final loop in 6 ML and data reports, 2 since 2024; details are sparse. The company's interview-prep page describes greenfield coding and refactoring rounds for software engineers.

05Final loop: ML depth & applied design

30-60 min discussion rounds, sometimes at a whiteboard; may be folded into the hiring-manager screen for applied scientists

Interviewers probe how you would model a real Zillow problem: predicting sale price, handling rural versus urban homes, removing outliers, filling missing prices from nearby sales and validating the result. Recent prompts move to search systems, multimodal and generative models, and reasoning versus standard LLMs, and one applied scientist was given an application problem to design an algorithm for. Expect follow-up on cross-validation, hyperparameter search and metrics.

What it tests

  • Framing a housing or marketplace problem as an ML task
  • Validation design, hyperparameter search and metric choice
  • Classical methods such as KNN and regression, explained from scratch
  • Search, multimodal and LLM system trade-offs
  • Improving an existing production model

How to prepare

  • Read Zillow's public posts about the Zestimate and its models; candidates were asked how they would improve them.
  • For any design prompt, cover data, features, model, offline metric, online test and failure cases in that order.
  • Prepare a short comparison of reasoning-tuned and standard chat models, and of CNNs and vision transformers.

Free reported questions for this stage

ML modeling or design rounds after the screen in 6 ML and applied science reports, 3 since 2024 (search design, multimodal and LLM topics); one 2024 hiring-manager screen also had a design question.

06Final loop: product & business case (data science)

30-60 min conversational case, often with a data science manager

Data science candidates describe open business cases: investigate why saved homes fell year over year, improve a maps product, or plan a real-estate analytics project using statistics. Questions about Zillow's business model also recur. Almost all of this evidence predates 2024, so confirm whether your loop has a case round.

What it tests

  • Breaking down a metric drop into causes you can test
  • Proposing product improvements and how to measure them
  • Knowledge of Zillow's business and customers

How to prepare

  • Use a fixed structure for metric drops: definition and data checks, segments, external factors, then hypotheses.
  • Learn how Zillow makes money and who its customers are before the loop.

Product or business case questions in 6 data scientist reports from 2017 to 2023, in the screen or the final loop; none in ML engineer or applied scientist reports.

07Behavioral & project deep dive

One round in the final loop, plus segments of most screens; STAR-style questions

Expect a detailed walk-through of a recent project: goal, your role, why you chose a model, its pros and cons and the outcome, followed by standard questions about mistakes, deadlines, feedback and teamwork. The company's preparation page recommends the STAR method and includes an example for technical candidates.

What it tests

  • Ownership and depth on one end-to-end project
  • Handling mistakes, deadlines and feedback
  • How you present results to stakeholders
  • Motivation for Zillow

How to prepare

  • Prepare one project you can defend in depth, including what you would do differently.
  • Structure every story with situation, task, action and result, as the company's own guide suggests.
  • Have concrete numbers ready for impact; interviewers ask about outcomes.

Behavioral or project deep-dive questions in 16 ML, data and applied science reports (4 since 2024) plus one 2024 engineering report. The company's interview-prep page covers the STAR method.

Questions and themes reported most often

  • Take-home: predict house sale prices from a provided dataset (7 reports, take-home modeling project (older loops))
  • SQL on users, states and saved homes (7 reports, technical or hiring-manager screen)
  • Deep dive on a past project: model choice, trade-offs, outcome (7 reports, behavioral & project deep dive)
  • Easy-to-medium data-structure coding problem (6 reports, technical or hiring-manager screen)
  • Overfitting, regularization and linear regression basics (5 reports, technical or hiring-manager screen)
  • Why Zillow and how its business works (4 reports, behavioral & project deep dive)
  • Explain CNNs, pooling, attention and vision transformers (3 reports, technical or hiring-manager screen)
  • LLM application questions: hallucination, retrieval chunking, agent loops (2 reports, technical or hiring-manager screen)

What Zillow asks about

  • ML concepts & modeling30%
  • Coding & data structures22%
  • SQL & data analysis14%
  • Behavioral & project deep dive14%
  • Take-home modeling project12%
  • Product & business cases8%

Zillow interview prep checklist

  • Built and wrote up a house-price model with sensible validation and clear figures
  • Solved several easy-to-medium heap, stack and tree problems with tests in under 30 minutes each
  • Wrote SQL for share of users by state and average time to first action
  • Can explain overfitting, L1 versus L2, attention and CNN pooling in plain words
  • Sketched a ReAct-style agent loop with a step budget and stop condition
  • Prepared a STAR-structured deep dive on one end-to-end project
  • Asked the recruiter for the format and focus of every round

Tips

  • Ask your recruiter what each round covers; several candidates were told one focus and tested on another.
  • For ML engineer roles, keep standard data-structure problems sharp and practise explaining a second approach.
  • For applied scientist roles, review regression, overfitting, CNNs and attention until you can explain them without notes.
  • For data scientist roles, drill SQL anti-joins, percentages by group and date differences on user-activity tables.
  • Know the Zestimate and how home prices vary by location; housing-specific modeling questions recur.
  • If you get a take-home, invest in the written report and figures, not only model accuracy.
  • Prepare for LLM application questions (hallucination, retrieval, agents) if the team builds AI features.
  • Expect slow scheduling and little feedback; keep other processes moving in parallel.

See the rest of the Zillow loop

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

Unlock with Pro

Practice the Zillow loop

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

Zillow interview FAQ

How many interview rounds does Zillow have for ML roles?

Candidates report 7 stages: Recruiter or HR call, Take-home modeling project (older loops), Technical or hiring-manager screen, Final loop: coding, Final loop: ML depth & applied design, Final loop: product & business case (data science) and Behavioral & project deep dive. A recruiter or HR call, then one technical screen of about an hour, often with the hiring manager. Before about 2023 most data and applied science candidates also did a take-home modeling project, usually house-price prediction, before or after that screen; no report since 2023 mentions one. The final loop is about four or five interviews of 30-60 minutes, now held remotely: one 2024 ML engineer loop was three coding rounds and one behavioral round, while data science loops mixed SQL or R coding, business cases and behavioral rounds with managers from several teams. Applied scientist reports since 2024 mostly stop at the hiring-manager screen, so their final loop is thinly documented. Several candidates describe slow scheduling and little feedback after a rejection, so follow up on your own.

How long does the Zillow ML interview process take?

Reported timelines run 2-8 weeks.

What does Zillow ask in ML interviews?

By reported question mix: ML concepts & modeling 30%; Coding & data structures 22%; SQL & data analysis 14%; Behavioral & project deep dive 14%; Take-home modeling project 12%; Product & business cases 8%. The most repeated themes: Take-home: predict house sale prices from a provided dataset; SQL on users, states and saved homes; Deep dive on a past project: model choice, trade-offs, outcome.

How should I prepare for the Zillow ML interview?

Ask your recruiter what each round covers; several candidates were told one focus and tested on another. For ML engineer roles, keep standard data-structure problems sharp and practise explaining a second approach. For applied scientist roles, review regression, overfitting, CNNs and attention until you can explain them without notes.

Are there free Zillow interview questions?

Yes. LeetML has 13 questions reported from Zillow interviews, and 5 of them are free to solve in the browser; they are listed by stage above.

Related interview guides

Official sources