Canva ML Engineer interviews, as candidates describe them, combine a recruiter screen, hands-on or AI-assisted coding, and in some loops several back-to-back interviews. Topics include search fairness, classical ML from scratch, and a timed online assessment.
$165K
Avg. Base Comp
$200K
Avg. Total Comp
2-6 rounds
Typical Rounds
14 days
Process Length
Candidates describe Canva's ML Engineer interview as more hands-on than conceptual. Two candidates report a recruiter screen followed by an AI-assisted pair coding round. One of them was told to bring their own development environment with unit testing and to use Cursor or GitHub Copilot, with prompting evaluated. Both describe a photo-search gender diversity problem built on a skewed content library. In one candidate's version, queries such as bodybuilder returned mostly male images, and the ask was a general fix that could ship quickly. That candidate was steered away from retraining or rebalancing data toward non-ML mitigations, such as round-robin style indexing and diversity measures like the Shannon index. The other reports reranking an M M F F X result set, then reading a YAML or CSV file and comparing output with expert rankings using a frequency-based distance such as Hamming distance.
For senior candidates, one 2024 report describes a phone screen, three pair coding sessions, an ML system design round and a behavioral interview. Coding there meant implementing k-means in native Python, alongside questions on confusion matrices and training PCA, with an emphasis on building implementations rather than discussing them conceptually. Another candidate reports fill-in-the-missing-function coding under time pressure and a four-interview morning that included business-case discussion. A separate applicant describes a timed HackerRank assessment mixing algorithms, probability, statistics and multiple-choice ML questions, with a very high bar.
In short, practice writing clean, testable Python quickly, and rehearse explaining trade-offs aloud. Also review search ranking fairness and diversity-aware retrieval. Four of the five reports ended without an offer, and the fifth was still waiting for a reply, so expect a demanding bar.
Synthesized from 5 candidate reports by our editorial team.
Had an interview recently?
Share your experience. Unlock the full guide.
Real interview reports from people who went through the Canva process.
Interviewer was very helpful.
He explained the question really well and we went through it together, and gave hints. He said I needed a working solution by the end of the interview.
There were 2 rounds and it was very much what I expected. The HR round had screening questions and past-work discussions. In the AI-assisted coding round, I was given a question and we had to build a working, visualizable solution.
The question concerned gender diversity in photo-search results when the content library was heavily skewed. It had two parts: rerank an M M F F X result set to be more diverse, then read a YAML/CSV file, run the code, and compare output with expert rankings using a frequency-based distance metric such as Hamming distance.
Prep tip from this candidate
Practice explaining a practical diversity-aware reranking approach, then implement and compare it against supplied rankings in a local development environment.
Share your own interview experience to unlock all reports, or subscribe for full access.
Sourced from candidate reports and verified by our team.
Topics based on recent interview experiences.
Featured question at Canva
Write a query to get the percentage of search queries where all ratings are less than 3 rounded to two decimals
| Question | |
|---|---|
| Hurdles In Data Projects | |
| Duplicate Rows | |
| k-Means from Scratch | |
| Unified Event Pipeline | |
| Merge Sorted Lists | |
| Subscription Overlap | |
| String Shift | |
| First to Six | |
| P-value to a Layman | |
| Bagging vs Boosting | |
| Find Bigrams | |
| Job Recommendation | |
| The Brackets Problem | |
| Compute Deviation | |
| Find the Missing Number | |
| Level Of Rain Water In 2D Terrain | |
| Permutation Palindrome | |
| Scrambled Tickets | |
| Find the First Non-Repeating Character in a String | |
| 500 Cards | |
| Nearest Common Ancestor | |
| Minimum Change | |
| Raining in Seattle | |
| Same Algorithm Different Success | |
| Jars and Coins | |
| Type-ahead Search | |
| Lasso vs Ridge | |
| Amateur Performance | |
| Prime to N |
Synthesized from candidate reports. Individual experiences may vary.
Candidates report an opening call with HR covering the role, their background and past-work discussion. One candidate says the role was framed around the User Data Platform group, with NLP, fine-tuning and open-source LLM prompting. Candidates describe the recruiters as professional and helpful, though the later technical rounds were less predictable.
One applicant reports a HackerRank link with a two-hour window and roughly four to eight questions. They included algorithms, scripts to complete, probability, statistics and multiple-choice ML items. The candidate says near-perfect scores appeared to be expected. This step was not mentioned by the other candidates, so it may not apply to every loop.
Two recent candidates describe an AI-assisted coding round focused on producing a working solution. One reports a one-hour session in their own setup, with tools such as Cursor or Copilot, and an emphasis on clean production code. The reported task was a diversity-aware photo-search problem, and one candidate describes reranking a result set and comparing it against expert rankings. Interviewers reportedly gave hints, and one candidate was pushed toward non-ML mitigations.
A senior candidate in 2024 reports pair coding sessions that required k-means in native Python, plus questions on confusion matrices and training PCA. Another reports filling in missing functions across multiple problems under time pressure. Candidates may be expected to show hands-on fundamentals rather than only discuss ML concepts, even at senior level.
Some candidates report several interviews in one day or morning. One describes four 45-minute rounds, and another describes four consecutive interviews. Reported content includes ML system design, behavioral discussion and business-case problems. The format varied between reports, so confirm the schedule with your recruiter. Prepare for fatigue and for explaining your reasoning clearly.
