OpenAI is at heart a research lab, and their growth into a valuable business was mostly a product of striking gold in the research department. I suppose their process reflects this.

I'll echo the poster mentioning that each interview took 2+ weeks to schedule. I entered the loop in early November and finally took my initial screeners in early February. During that time, I turned down two offers elsewhere because I was still waiting for their interview loop to grind along.

Theoretically, these screeners were a system design interview and a management roleplay. I was prepped that if I could architect a service such as Yelp or Twitter at scale, I'd have no issues with the system design interview. The roleplay was explained straightforwardly but is more a judgment on style than anything, as there are always multiple ways to deal with these challenges and no obvious "right" answer. I am a very seasoned manager at this point in my career and have confidence in the way I dealt with their management scenarios.

My issue was with the "system design" interview. The first 5 minutes were what you would expect, with selecting database systems, designing a schema, SQL vs. NoSQL, API layer, cloud scaling, load balancing, etc. Easy. I breezed through it.

Once that was completed to the interviewer's satisfaction, things started to go way off course. The interviewer apparently exhausted the system design question and then decided to zoom in on one tiny aspect of the design, a geospatial index, and start asking heavy algorithmic questions: how can I optimally partition up the space? What is the time complexity of an R-tree vs. quadtree solution to this problem? Design the algorithm for traversing the tree. What is the stopping criterion? No Googling allowed.

Since I haven't dealt with these data structures directly in over 15 years, I was not able to regurgitate the algorithm for constructing an R-tree perfectly from memory. The interviewer also kept adding and removing constraints (e.g., "now do it without the radius parameter" and then three questions later the interviewer asked why I didn't use the radius parameter without mentioning it was back on the table).

I feel disappointed that my chances of working here as a manager, not even an engineer, were ruined by an interview that should not have even been on the loop to begin with. And having to turn down other offers because they couldn't get their process to move at a reasonable speed made it even worse.

Lots and lots of low-level geospatial indexing questions.