Canva ML Engineer Interview Guide

Canva ML Engineer candidates should prepare for an interview that can move beyond conventional model-building into practical search-ranking mitigation. One candidate described an HR conversation focused on screening and past work, followed by an AI-assisted coding session where the interviewer expected a working, visualizable solution by the end. The core prompt concerned photo-search results whose underlying image library was heavily skewed by gender. Rather than relying on manual result elevation for individual queries, the task was to propose a general mitigation.

Be ready to explain how you would rerank a small result list for greater diversity, then turn that thinking into code. The reported exercise also included reading YAML or CSV input, running the solution, and comparing generated rankings with expert rankings using a distance measure such as Hamming distance. A separate senior-level candidate described the same fairness scenario and said the discussion was steered toward non-ML approaches, including round-robin-style ordering and diversity metrics.

For the HR conversation, have concise examples of past work ready. For the coding session, practice making tradeoffs explicit: what the reranker optimizes, how you would evaluate diversity against relevance, and how you would produce a runnable result under time pressure. Reports are limited and show different process lengths and totals, so the exact sequence may vary.