Canva Data Engineer Interview Guide

The interview process had three main stages.

The first was a recruiter call covering my background, motivation for joining Canva and general fit for the role.

The second stage was a live SQL interview conducted in a Google Doc. There was one main problem with multiple parts. I was given several input tables and asked to write queries that produced a specified output. The SQL itself—mainly joins and filters—was not especially difficult. The real challenge was understanding the schema, identifying the grain of each table and clarifying how the data related before writing the query. A later part involved writing SQL to ingest data into a target table. After passing the SQL round, I progressed to a four-session final loop: data modelling, system design, AI-assisted coding and behavioural.

The data-modelling round started with a fairly complex schema containing a lot of information, much of which was irrelevant to the problem. The first challenge was filtering out that noise and identifying the tables and fields that actually mattered. The modelling itself was reasonable and focused largely on applying SCD Type 2 and adapting an existing model as the business requirements changed.

System design was where I started sweating. The prompt involved designing a data platform capable of operating at extremely large scale. It was very broad, and there were many valid directions the discussion could take. In hindsight, I should have spent more time clarifying which areas the interviewer wanted me to explore in depth. I covered scalability and platform architecture, but I felt that I missed some topics they were looking for, particularly data contracts and the mechanisms teams would use to publish and consume data in a data-mesh architecture.

In the AI assisted coding interview, you are not expected to write a single line of code. Instead, you guide an AI coding assistant through the entire development process. My task was to retrieve data from a set of dummy APIs. The interviewer was interested in how I clarified requirements, broke the problem into smaller steps, prompted the AI, reviewed its output and corrected it when necessary.

The behavioural round was fairly conventional. Questions included describing a challenging project, explaining why it was difficult and discussing what I did to overcome the problems. The interviewer specifically asked me to structure my answers using the STAR method, so having a few detailed examples prepared was useful.

Questions asked: The SQL round had one problem with multiple parts. I was given several tables and asked to produce a specified output using joins and filters. The main challenge was identifying each table’s grain and understanding the relationships. I was then asked to write a query to ingest the results into a target table.

In data modelling, I received a complex schema containing a lot of irrelevant information. I had to identify what mattered, extend an existing model and adapt it as the requirements changed. This included applying SCD Type 2.

The system-design prompt was to design an event-data platform that could operate at extremely large scale. The discussion covered ingestion, processing, storage, scalability, latency and failure handling. It also touched on data contracts and how teams publish and consume data under a data-mesh architecture.

For AI-assisted coding, I was given documentation for two dummy APIs and asked to build a Python solution to extract and load data from them. The APIs included pagination, sorting and ascending/descending options. I was expected to guide the AI through the implementation, review its code and tests, and correct it when necessary—not write the solution manually.

The behavioural round included questions such as: “Tell me about a challenging project. Why was it challenging, and what did you do to overcome it?” The interviewer specifically requested answers using the STAR structure.