Hey guys, just wanted to give back to the community and help anyone else in the process.

Rippling has a small question pool and the questions are widely available on leetcode discuss section. I, like many others, got the driver-delivery system problem.

They care a lot about the money/cost datatype. Even though I knew we are supposed to pick BigDecimal and not float/double, he asked for the detailed reason why. Furthermore, when I was in the middle of the coding round, Copilot added 28 precision bits on the number and the interviewer asked to change that since the representation is meant to be like real money. So I reduced the precision to 4 bits, in which case you have a dollar value and 2 digits after the decimal point.

When running the test cases, some value was meant to be zero but the code was returning 0.005. The interviewer flagged this as well. Long story short, Copilot was not helping me accurately set the precision (even with quantize in Python).

Be careful with AI coding. Copilot started to add unnecessary functions and even some lines of unoptimized code (which is not a good place to be in). So it is important to clarify to the interviewer that you are aware what Copilot spit out is not accurate and that you are going to change it.

My interviewer seemed to be more interested in jumping directly into coding instead of me completely walking through the solution first, and this was apparent throughout (this might be different for others, so clarify how you want to approach the solution).

I coded up the initial part (3 functions) and then he gave me some test cases to run in my local system. Then he added 2 follow-up functions (pay_upto_time and get_total_unpaid_cost). Then he gave me test cases for the follow-up question.

Here again, he asked me if I knew how real financial systems deal with decimal numbers and rounding up/down rules. He then provided the final follow-up and said, "even though we don't have enough time, walk me through your solution." This was the analytics part of get_maximum_simultaneous_drivers_in_last_24hrs().

Then it was some QnA. During the coding, he also acknowledged that Copilot might be slowing me down instead of actually helping me speed things up, so again, be comfortable with the tool you are using.

We wrapped up by pasting the code in HackerRank (I guess they do this for audit). Overall, I think it went pretty well, hoping for a call in a few days.