Company interview guide
Canva Machine Learning Interview Guide (2026)
An AI-assisted applied ML problem, ML coding without AI, ML design and a values-based behavioral round.
Canva's machine learning interview loop has 7 reported stages, from recruiter call to behavioral and values; typical length: 2-5 weeks.
Roles covered: ML Engineer and Data Scientist. The question bank holds 2 questions reported from Canva interviews, 1 of them free to solve in your browser.
- Interview process length
- 2-5 weeks
- Stages
- 7
- Reports linked to bank questions
- 2
- Bank report dates
- Mar 2025–Jun 2026
- Questions in bank
- 2 (1 free)
- Last reviewed
- October 2026
How the Canva interview loop works
The hiring page lists a recruiter call (with technical questions for engineers), a peer technical interview with a problem-solving challenge, and a final interview of about three hours; it quotes 4-5 weeks to a decision, and candidates report anywhere from two weeks to longer waits when interview slots were scarce. For ML engineers, reports since 2025 run: recruiter screen, a one-hour AI-assisted coding round on an applied ML business problem, then a final block of three or four interviews of 45-60 minutes: ML coding or pair programming without AI, ML system design (one 2026 report adds an ML system review), and behavioral, sometimes with a chat with the engineering manager. Reports from 2023-2024 describe a timed online test for some roles, a pair-coding screen on existing ML code, and a final morning of several coding rounds plus design and behavioral. Data science loops swap the coding rounds for SQL, Python or R and statistics. Engineering roles have no take-home in the current process.
The Canva interview process, stage by stage
01Recruiter call
About 30 min video call with a recruiter; engineering candidates also get some technical questions
A talk about your experience, motivation and the role. Candidates mention being asked why Canva, about past projects and experimentation, about the tools and editor they use, and in one senior ML case whether they had worked with terabyte-scale data pipelines. Recruiters were reported to be clear about compensation early.
What it tests
- Why Canva and why this team
- Fit between your past projects and the role
- Familiarity with the team's stack and data scale
- Logistics: level, location, timeline
How to prepare
- Know the product: candidates were asked about Canva itself, not just the role.
- Ask what each later round looks like, including which ones allow AI tools and what to install beforehand.
- Have one project ready that shows scale or experimentation, since both came up here.
A recruiter or HR screen opens the loop in 17 ML and data science reports (15 since 2024). The company's hiring page lists this call and says engineers also get technical questions.
02Online test (some roles, 2023-2024)
Timed, unproctored online test of 1-2 hours: coding problems plus multiple-choice probability, statistics and ML questions; data science versions use pandas and SQL
Several 2023-2024 reports, including graduate and mid-level ML roles, describe an automated test before any live interview. ML versions mixed easy and medium algorithm problems, partly written scripts to finish, and short questions on Gaussian distributions and other basics; data science versions tested pandas manipulation and SQL. No report since 2025 mentions it, so ask your recruiter whether it still applies.
What it tests
- Easy and medium algorithm problems under time pressure
- Completing partly written Python scripts
- Probability and statistics basics such as Gaussian properties
- pandas and SQL data manipulation (data science)
How to prepare
- Treat it as a speed test: one candidate felt near-perfect scores were needed to pass.
- Practise writing pandas and SQL that runs first time, since there is little time to debug.
Described in 6 reports from 2023-2024 (four ML, two data science); none of the 2025-2026 ML reports mention it, and the hiring page does not list it.
03AI-assisted coding round
About 60 min in your own editor with an AI assistant (Copilot, Cursor, Claude or similar) and a test framework installed in advance
Since 2025 this is the main ML technical screen. You get a short, open-ended business problem and are expected to build a clean, tested solution with AI help while the interviewer watches how you scope it and review the output. Most reported prompts are about bias in search or recommendations, such as image results for a query skewed heavily toward one gender, followed by how to measure whether the fix worked. The company says it assesses problem breakdown, ownership of the code and judgment in using AI, not tool tricks.
What it tests
- Scoping a vague, time-boxed business problem into something shippable
- A general mitigation such as a re-ranking step, and how to evaluate it against ground truth
- Clean project structure and unit tests
- Reviewing, correcting and explaining AI-generated code
How to prepare
- Set up a project skeleton with tests and your assistant before the call; candidates were told both must be installed.
- Ask clarifying questions and state your own plan before prompting the AI.
- Read every generated line and be ready to justify it; over-delegating is a named failure mode.
- Prepare a fairness or diversity re-ranking approach and a diversity metric for the top results.
Free reported questions for this stage
- Gender Diversity in Image Search ResultsML System DesignMediumSolve
An AI-assisted round after the recruiter call in 6 ML reports since 2025; 5 got a search or recommendation bias problem. The company's engineering blog says ML candidates must use AI tools.
04ML coding without AI (pair programming)
One or more 45-60 min pair-programming sessions in Python; earlier the screen, now part of the final block
Candidates repeatedly describe implementing classic algorithms in plain Python, most often k-means or another clustering method, along with a confusion matrix or small evaluation script, PCA, or filling in missing functions in an existing ML codebase. Before 2025 this was the technical screen and the final morning often had two or three such sessions; since 2025 one coding round without AI sits in the final block. No reports describe standard algorithm puzzles in this round.
What it tests
- k-means or clustering from scratch without libraries
- Evaluation code: confusion matrix and derived metrics
- PCA and other classic methods explained and coded
- Reading existing code and completing it rather than rewriting it
How to prepare
- Spend the first minutes reading the provided code; one candidate lost time by starting from scratch.
- Be able to write k-means, a confusion matrix and PCA in plain Python and numpy from memory.
- Practise without autocomplete or AI so the switch from the previous round does not throw you.
ML coding or pair programming in 12 ML reports (11 since 2024); clustering, confusion matrices or completing existing code recur in seven. The hiring page describes a peer technical interview and technical problem-solving in the final.
05ML system design
About 45-60 min discussion in the final block; one 2026 report had a separate ML system review round
Design an ML system for a Canva-style business problem. The reported prompts include a classification system that routes uncertain cases to human reviewers and an open technical discussion of a business case. One senior candidate also had a round reviewing an existing ML system, and an engineering manager described the team as mostly software engineering with little model training.
What it tests
- Framing a business problem as an ML task with clear metrics
- Classification with a human-in-the-loop review step
- Data pipelines at large scale
- Critiquing an existing design and its trade-offs
How to prepare
- Practise a moderation-style classifier with thresholds, human review and feedback loops end to end.
- Bring the production side: serving, monitoring and data pipelines, not only the model.
An ML design or technical business-case round in 7 ML reports since 2024, consistently in the final block; prompts are only given in two of them.
06Data science track: SQL, Python and statistics
Live rounds over a shared screen, sometimes in a plain notepad rather than an IDE: SQL first, then Python or R and statistics
Data scientists report a SQL round on a given table with questions that get harder, sometimes with a short theory question, followed in some loops by Python or R and statistics rounds. Senior candidates also met dashboard design and metrics questions about Canva users. The hiring page describes a take-home brief for non-engineering roles, but only one old data science report mentions one.
What it tests
- SQL on a described table, from simple to multi-step queries
- Metrics for measuring product and user performance
- Dashboard design for stakeholders
- Python or R and statistics fundamentals
How to prepare
- Practise writing SQL without an editor or autocomplete.
- Prepare metric definitions for a freemium design product: activation, retention, conversion to paid.
A live technical round in 6 data science reports since 2023, four of them SQL-focused; the stats and Python rounds and dashboard questions each appear in only one or two.
07Behavioral and values
About 45-60 min in the final block, scenario questions based on past experience; sometimes a separate chat with the engineering manager
Questions about real situations from your work: successes, difficulties, collaboration and stakeholder management, scored against the company's skills framework (craft, strategy, communication, leadership and coaching). Candidates also report why Canva questions and a culture-fit framing.
What it tests
- Specific stories of handling difficulty and conflict
- Stakeholder management and leadership for senior roles
- Motivation for Canva and its product
- Communication of technical work to others
How to prepare
- Map two or three stories to each of craft, strategy, communication and leadership.
- Expect each round to have a single interviewer, so make your reasoning explicit.
A behavioral or culture-fit round in 10 ML and data science reports since 2024, always in the final stage. The hiring page says the final includes experience-based questions tied to its skills framework.
Questions and themes reported most often
- Mitigate gender bias in image search or recommendations, then evaluate the fix (5 reports, ai-assisted coding round)
- Implement k-means or another clustering algorithm from scratch (4 reports, ml coding without ai (pair programming))
- Complete missing functions in an existing ML codebase (4 reports, ml coding without ai (pair programming))
- SQL questions of rising difficulty on a given user table (4 reports, data science track: sql, python and statistics)
- Online test: coding plus probability and statistics questions (4 reports, online test (some roles, 2023-2024))
- Confusion matrix and a basic evaluation script (3 reports, ml coding without ai (pair programming))
- Why Canva, and knowledge of the product (2 reports, recruiter call)
- Classification system with a human review step (1 reports, ml system design)
What Canva asks about
- ML coding without AI30%
- AI-assisted applied ML problem20%
- ML system design18%
- SQL & product analytics12%
- Behavioral & values12%
- Probability, stats & algorithms tests8%
Canva interview prep checklist
- Built a small tested project with an AI assistant and can explain every line it wrote
- Wrote k-means from scratch in plain Python and evaluated it
- Implemented a confusion matrix and the metrics derived from it without libraries
- Sketched a diversity re-ranking for skewed search results and a way to measure it
- Designed a classifier with a human review loop, including thresholds and monitoring
- Prepared stories for each part of the skills framework and a clear answer to why Canva
Tips
- Ask the recruiter which rounds allow AI tools and which do not; the current ML loop has both.
- Rehearse an end-to-end AI-assisted session: skeleton project, tests, prompting in small steps and reviewing every change.
- Write k-means, a confusion matrix with precision and recall, and PCA from memory in plain Python.
- When given existing code, read it first and fill the gaps instead of rewriting it.
- Prepare a fairness-aware re-ranking idea and a diversity metric; bias in search is the most reported prompt.
- Expect a long final block of three or four back-to-back interviews and plan your energy.
- Learn the product and the skills framework (craft, strategy, communication, leadership) before the final.
- Data science candidates: practise SQL without an IDE and product metrics for a freemium tool.
See the rest of the Canva loop
The later stages, the questions reported for them and how to prepare are in Pro.
Unlock with ProPractice the Canva loop
Work the stages in order: every reported question for this company, grouped by round, with solutions in the app.
Canva interview FAQ
How many interview rounds does Canva have for ML roles?
Candidates report 7 stages: Recruiter call, Online test (some roles, 2023-2024), AI-assisted coding round, ML coding without AI (pair programming), ML system design, Data science track: SQL, Python and statistics and Behavioral and values. The hiring page lists a recruiter call (with technical questions for engineers), a peer technical interview with a problem-solving challenge, and a final interview of about three hours; it quotes 4-5 weeks to a decision, and candidates report anywhere from two weeks to longer waits when interview slots were scarce. For ML engineers, reports since 2025 run: recruiter screen, a one-hour AI-assisted coding round on an applied ML business problem, then a final block of three or four interviews of 45-60 minutes: ML coding or pair programming without AI, ML system design (one 2026 report adds an ML system review), and behavioral, sometimes with a chat with the engineering manager. Reports from 2023-2024 describe a timed online test for some roles, a pair-coding screen on existing ML code, and a final morning of several coding rounds plus design and behavioral. Data science loops swap the coding rounds for SQL, Python or R and statistics. Engineering roles have no take-home in the current process.
How long does the Canva ML interview process take?
Reported timelines run 2-5 weeks.
What does Canva ask in ML interviews?
By reported question mix: ML coding without AI 30%; AI-assisted applied ML problem 20%; ML system design 18%; SQL & product analytics 12%; Behavioral & values 12%; Probability, stats & algorithms tests 8%. The most repeated themes: Mitigate gender bias in image search or recommendations, then evaluate the fix; Implement k-means or another clustering algorithm from scratch; Complete missing functions in an existing ML codebase.
How should I prepare for the Canva ML interview?
Ask the recruiter which rounds allow AI tools and which do not; the current ML loop has both. Rehearse an end-to-end AI-assisted session: skeleton project, tests, prompting in small steps and reviewing every change. Write k-means, a confusion matrix with precision and recall, and PCA from memory in plain Python.
Are there free Canva interview questions?
Yes. LeetML has 2 questions reported from Canva interviews, and 1 of them are free to solve in the browser; they are listed by stage above.
Related interview guides
- AtlassianAn ML craft deep dive first, then practical coding, ML system design for Jira and Confluence, and a values round that carries real weight.
- AdobeMedium-level algorithm coding, practical ML fundamentals and deep project discussion, with a job talk for research roles.
- PinterestRapid-fire ML fundamentals, edge-case-heavy coding and embedding-centric recommendation design.
- AirbnbMarketplace ranking design, a hard look at your past models, one practical coding round and a values check.
- SpotifyPractical Python and SQL, music-flavoured modeling cases and recommendation design, with a real values round.