The loop was a founder screen followed by a coding question and panel. Everyone was friendly for the founder screen, although a great deal of system design and strategic Q&A was present. The interviewers seemed *really* unprepared for the coding screen. Three people jumped on and off with no introduction and the person who ultimately did the interview wasn't the one listed on the invite, or even anyone I was able to find in the company's LinkedIn profile. They started 5 min late and didn't apologize for it. There was no coderpad and I was forced to set up the IDE, create a virtualenv, and download 50+ GB of models and libraries during the next 10 min while the interviewer looked on. I implemented the solution in the remaining time, constantly getting interrupted by the interviewer with questions like "where is the kNN implemented?" because I used an in-memory vector database rather than an explicit call to sklearn.Neighbors. It was clear the interviewer didn't understand a number of details of the solution, including the vector DB's ANN algorithm functioning much faster than naive kNN. I was stopped for Q&A just as I was outputting the confusion matrix (basically a cosmetic display of the results), with about 12 min left in the interview. I later discovered that they marked my response correct but incomplete. Given the general lack of organization, wasting almost half of the interview time, and marking the whole solution incomplete because half of a print statement was missing and the interviewer didn't clarify that the last 12 min were going to be unavailable upfront, it was a pretty negative experience.