In the code review round, I was given a pull request for an issue where a Python application was crashing under heavy load. The pull request changed the existing synchronous processor, which handled one request at a time, into a batched processor that handled 40 requests at a time. The repository still included the synchronous processor as a reference. However, if creation of the batched processor failed, the new code returned None instead of falling back to the synchronous processor. I pointed that out.
The pull request description also said that the new batched processor could handle 6,000 requests per second per instance, while the peak load stated in the README was 8,000 requests per second per instance.
The pull request was huge, and its description and README files were also huge. The expectation was to use AI for everything, and the interviewers judged how well I used it. Trying to work manually would make it impossible to manage the time. There were many bugs in the pull request, so I needed to prioritize them and identify the critical ones with AI. I also needed to remember to validate the AI's suggestions.
Discussion
Loading comments…