Company interview guide
Unity Machine Learning Interview Guide (2026)
Recruiter and hiring-manager screens, often a take-home on real ads data, then conversational technical rounds on ML depth and systems.
Unity's machine learning interview loop has 6 reported stages (Recruiter call, Hiring manager screen, Skills assessment: take-home or coding test, Coding rounds, Final loop: ML depth, systems & take-home review, Behavioral, stakeholders & hiring manager wrap-up); typical length: About 4 weeks.
Roles covered: ML Engineer and Data Scientist. The question bank holds 1 question reported from Unity interviews, 1 of them free to solve in your browser.
- Interview process length
- About 4 weeks
- Stages
- 6
- Reports linked to bank questions
- 1
- Bank report dates
- Apr 2022–Apr 2022
- Questions in bank
- 1 (1 free)
- Last reviewed
- October 2026
How the Unity interview loop works
A roughly 30-minute recruiter call, a 30-60 minute hiring-manager screen, a skills assessment for many roles, then a final loop of about three to five interviews of 30-60 minutes, done over video and either on one day or spread out. Data science reports from 2019 to 2024 most often name a take-home modelling task with one to two weeks to finish; a few ML roles got a timed online coding test instead. Recent ML engineer reports describe the final loop as technical deep dives with individual contributors, sometimes two coding and two behavioral rounds, and a closing check-in with the hiring manager. The company says you get an agenda for the final stage beforehand and that every candidate for a role gets the same questions. Rounds vary by team and few recent reports give detail.
The Unity interview process, stage by stage
01Recruiter call
About 30 min by phone or video with a recruiter or talent sourcer
A friendly intro to the role and team, your background and why you want to move. Candidates repeatedly praise the recruiters, who often stay with them through the whole process and share preparation notes for each round.
What it tests
- Why Unity, and why this team
- Your background and the technologies you have used
- Motivation for a change, such as moving from academia to industry
- Logistics: location, level, timeline
How to prepare
- Ask the recruiter for the type and length of every later round and whether a take-home is part of it.
- Have a two-minute summary of the project most relevant to the team ready.
- Unity does not interview by email or text, so treat such contact as a scam.
A recruiter or HR first call in 25 data science and ML reports, five of them from 2024-2026, usually about 30 minutes. The company's hiring page lists this phone screen as the first step.
02Hiring manager screen
30-60 min video call with the hiring manager or team lead; some teams send an engineer or senior scientist instead
Usually the second step and often technical: the manager asks about past ML work, model choices and the stack the team uses, plus some behavioral questions. In a few reports this screen was a one-hour conceptual ML interview, with SQL in one older data science loop.
What it tests
- Depth behind the projects on your resume
- ML fundamentals and how you would model a problem the team faces
- Fit with the team's stack, for example PyTorch, Spark or Kubernetes
- Behavioral or situational questions about past experience
How to prepare
- Prepare one project to walk through end to end, including why you chose each design.
- Read the job post closely and be ready to say where you have used each listed tool.
- Use the questions you are allowed to ask to learn what the take-home or final rounds will cover.
A hiring-manager or technical screen after the recruiter in 17 reports, five from 2024-2026; most call it partly technical. The company's hiring page describes a 30-60 minute hiring-manager screen.
03Skills assessment: take-home or coding test
Data science: a take-home modelling task on a subset of real data, usually one to two weeks; some ML roles: a timed online coding test (one report: 80 min, two tasks)
The take-home is the most distinctive part of older Unity data science loops: build a model for a business question such as conversion or ad-install prediction in a notebook, then defend it in the next rounds. Several candidates spent many hours on it, and a few complained about how it was reviewed. ML engineering roles have sometimes used short algorithmic coding tests instead, and recent ML engineer reports rarely mention either.
What it tests
- Framing a business question as a supervised learning problem
- Feature work, validation and handling missing values on messy real data
- Choosing and justifying a model and its loss
- Communicating results clearly in a notebook
- Easy-to-medium algorithm problems, for the coding-test variant
How to prepare
- Treat the notebook as a report: state the question, the baseline, the metric and why your model beats it.
- Expect to be asked what you would do differently, so note alternatives and their trade-offs as you go.
- Agree on how much time is expected before starting; some candidates spent a full working day.
- Keep the solution reproducible and simple enough to explain in a few minutes.
A take-home in 16 reports and an online coding test in 4, almost all from 2019-2023; only one 2024 data science report mentions it. The company's hiring page lists a role-specific skills assessment.
04Coding rounds
Live coding over video, about 45-60 min; reported for some ML engineer and older data science loops
Not every loop has one. A 2025 ML engineer loop had two coding rounds out of four virtual interviews, a 2026 loop had a coding assessment after the hiring-manager round, and older loops asked CS basics, SQL or a small ML utility such as k-fold cross-validation in Python.
What it tests
- Standard data structures and algorithms, such as dynamic programming and sorting
- Writing small ML utilities from scratch in Python
- SQL queries for data science roles
How to prepare
- Ask whether coding is live or an assessment, and which language and tools are allowed.
- Practise implementing cross-validation and simple metrics without libraries.
Live or scheduled coding rounds in five reports: two ML engineer loops from 2025-2026 and three loops from 2018-2021; most other reports describe no coding beyond the take-home.
05Final loop: ML depth, systems & take-home review
2-4 one-on-one video interviews of 30-60 min with data scientists, ML engineers and leads
The core of the loop. Interviewers dig into ML fundamentals, the architecture of systems you have built, and topics close to the team, such as recommender losses for ads or a 3D and computer-vision stack. Where there was a take-home, at least one round reviews it and asks for alternatives. Candidates describe these as open discussions rather than trick questions.
What it tests
- ML fundamentals: bias and variance, tree ensembles, regularization, missing data
- Designing or explaining an ML system end to end and justifying the choices
- Team-specific modelling, for example custom losses for recommendation or bidding
- Defending and improving your take-home solution
- Running models in production, including hyperparameter tuning
How to prepare
- Know tree ensembles well: random forest versus boosting and which loss each optimizes.
- Have a clear architecture diagram in your head for one system you built, with the trade-offs you made.
- Read about the team's product area (ads, monetization, 3D, vision) and connect your experience to it.
Free reported questions for this stage
- Random Forest vs Gradient Boosting (XGBoost)Classical MLMediumSolve
Technical ML, system design or take-home review rounds in 23 reports, five from 2024-2026; content is consistently discussion-based. The company's hiring page says the final stage asks job-specific questions.
06Behavioral, stakeholders & hiring manager wrap-up
1-2 conversations of 30-60 min with a product manager, engineering or data science manager, or director; recent loops end with a hiring-manager check-in
Unity weighs collaboration heavily. Many loops include a product manager or business stakeholder, and several end with a director or hiring-manager conversation. The company says behavioral questions follow its Principles and use the same set for every candidate.
What it tests
- Communicating with and influencing stakeholders
- Scoping a project and setting milestones for an ML component
- Teamwork, humility and reacting to others' work
- Alignment with the company's Principles
How to prepare
- Prepare STAR stories that map to empathy and respect, candor, urgency and the greater good.
- Have a story about convincing someone with data, and one about a project scoped with a product partner.
- Ask the hiring manager what success in the first six months looks like.
Behavioral, stakeholder or manager rounds in 21 reports, six from 2024-2026; many mix behavioral questions into technical rounds. The company's hiring page describes Principles-based behavioral questions.
Questions and themes reported most often
- Walk through an ML project you worked on and justify the design choices (10 reports, hiring manager screen)
- Take-home: build a prediction model on a real business dataset, often ads or conversion (7 reports, skills assessment: take-home or coding test)
- Explain the architecture of an ML system or design one (6 reports, final loop: ml depth, systems & take-home review)
- Defend your take-home: loss choice, alternatives and improvements (5 reports, final loop: ml depth, systems & take-home review)
- Behavioral: convincing someone, reacting to a colleague's faster work, why industry (3 reports, behavioral, stakeholders & hiring manager wrap-up)
- Tree ensembles: random forest versus boosting, overfitting and missing values (2 reports, final loop: ml depth, systems & take-home review)
- Experience with the team's tools, such as PyTorch, Spark, TensorFlow or Kubernetes (2 reports, hiring manager screen)
- Recommender and ads modelling: custom losses, reinforcement learning for bidding (2 reports, final loop: ml depth, systems & take-home review)
What Unity asks about
- ML depth & past projects30%
- Take-home modelling20%
- Behavioral & stakeholders20%
- ML system design15%
- Coding & SQL15%
Unity interview prep checklist
- Can walk through one ML project end to end, including what failed and why you chose each design
- Built a conversion or click prediction model in a notebook with a baseline, validation and a clear write-up
- Can explain random forest versus gradient boosting, their losses and how each overfits
- Can sketch a production ML system for ads or recommendations with its data flow and metrics
- Have four STAR stories mapped to the company's Principles
- Asked the recruiter which rounds your loop includes and whether there is coding
Tips
- Pick one or two projects and rehearse them end to end; nearly every round returns to them.
- If you get a take-home, write it as a clear report with a baseline, a metric and trade-offs, since the next round reviews it.
- Brush up on classical ML for tabular and ads data: tree ensembles, class imbalance, calibration and evaluation metrics.
- Be ready to sketch the architecture of a production ML system and explain why each piece is there.
- Learn the team's domain before the loop; ads ranking, bidding and 3D vision teams ask quite different questions.
- Prepare STAR stories against the company's four Principles.
- Ask the recruiter early whether your loop has a take-home, live coding or both; it differs by team and has changed over time.
- Expect friendly, discussion-style interviews and focus on explaining your reasoning out loud.
See the rest of the Unity loop
The later stages, the questions reported for them and how to prepare are in Pro.
Unlock with ProPractice the Unity loop
Work the stages in order: every reported question for this company, grouped by round, with solutions in the app.
Unity interview FAQ
How many interview rounds does Unity have for ML roles?
Candidates report 6 stages: Recruiter call, Hiring manager screen, Skills assessment: take-home or coding test, Coding rounds, Final loop: ML depth, systems & take-home review and Behavioral, stakeholders & hiring manager wrap-up. A roughly 30-minute recruiter call, a 30-60 minute hiring-manager screen, a skills assessment for many roles, then a final loop of about three to five interviews of 30-60 minutes, done over video and either on one day or spread out. Data science reports from 2019 to 2024 most often name a take-home modelling task with one to two weeks to finish; a few ML roles got a timed online coding test instead. Recent ML engineer reports describe the final loop as technical deep dives with individual contributors, sometimes two coding and two behavioral rounds, and a closing check-in with the hiring manager. The company says you get an agenda for the final stage beforehand and that every candidate for a role gets the same questions. Rounds vary by team and few recent reports give detail.
How long does the Unity ML interview process take?
Reported timelines run About 4 weeks.
What does Unity ask in ML interviews?
By reported question mix: ML depth & past projects 30%; Take-home modelling 20%; Behavioral & stakeholders 20%; ML system design 15%; Coding & SQL 15%. The most repeated themes: Walk through an ML project you worked on and justify the design choices; Take-home: build a prediction model on a real business dataset, often ads or conversion; Explain the architecture of an ML system or design one.
How should I prepare for the Unity ML interview?
Pick one or two projects and rehearse them end to end; nearly every round returns to them. If you get a take-home, write it as a clear report with a baseline, a metric and trade-offs, since the next round reviews it. Brush up on classical ML for tabular and ads data: tree ensembles, class imbalance, calibration and evaluation metrics.
Are there free Unity interview questions?
Yes. LeetML has 1 question reported from Unity interviews, and 1 of them are free to solve in the browser; they are listed by stage above.
Related interview guides
- RobloxGame-based assessments, practical coding screens, recommendation design and SQL-heavy data science loops.
- Electronic ArtsConversational loops built on ML fundamentals, past-project depth and game-data cases, with SQL for data scientists.
- SnapClassic coding plus fast, broad ML fundamentals and recommendation-heavy ML design, with values questions in every round.
- MetaFast classic coding, an AI-assisted coding project and product ML design for recommendations and integrity.
- NVIDIATeam-run loops that mix algorithm coding, deep learning built from scratch and GPU-aware LLM performance.