Goldman Sachs Fresher Interview Questions

“In a hackathon our team had to choose how to store user data with two hours left, and we did not know how many users the demo would have. Waiting for an answer would have used up the time. I chose the simplest option, a single database file, and wrote down that it was a guess. I also kept the storage code behind one small interface, so that swapping it later would take minutes, not hours. The demo worked and the guess was fine. What I kept was the habit of naming my assumption out loud and making the choice cheap to undo.”

“During my internship I presented a small analysis tool to a manager, and in the first minute she said the output was unreadable and she would not use it. It stung. I asked her to show me the first thing she looked for, and she pointed at a total buried at the bottom of the screen. I said I would come back the next day. I moved the totals to the top, cut the columns she never used, and returned. She used it from then on, and your result. I learned to ask what the person needs first, before the criticism becomes an argument.”

“I need to add orders, cancel them, read the best price and match. For each side I would keep a map sorted by price, so the highest buy and the lowest sell are found quickly. At each price there is a queue, because orders at the same price are served in the order they arrived. To cancel quickly, I would also keep a hash table from order id to its place in the queue. A heap gives the best price fast too, but removing a cancelled order from the middle of a heap is awkward, so I prefer the sorted map.”

“Floating point stores numbers in binary, and many simple decimals, like one tenth, cannot be written exactly that way. Each value is very slightly off. Add up many of them and the total drifts, and two amounts that should be equal can fail an equality check. For money I would store whole numbers of the smallest unit, such as cents, or use a decimal type that keeps exact decimal digits. I would also state the rounding rule clearly wherever division happens, such as splitting an amount among accounts, so the parts always add back to the whole.”

“The slow way compares every buy time with every later sell time, which grows with the square of the number of prices. The better way is one pass. As I walk through the prices, I remember the lowest price seen so far. At each new price, I work out the gain if I had bought at that lowest price and sold now, and keep the best gain. Then I update the lowest price if the current one is lower. If prices only fall all day, the best gain is zero and the answer is not to trade. Memory use stays constant.”

“I would first turn the prices into a list of daily changes, each as the change divided by the previous day's price. Then I would sort those changes from worst to best. If the question is the loss on a bad day that happens about once in a hundred days, I take the value at that rank in the sorted list. That needs one sort, which is quick even for years of data. I would be honest about the limit: the answer only describes the past, and a day worse than anything in the history can still happen. I would also check the list for missing days or data errors first.”

“I would tell my manager immediately, even before I am sure. I would say what the bug is, which report it touches and that I do not yet know the size. People may be making decisions on that figure, so an hour matters. Then I would run the corrected calculation beside the old one, to see which figures changed and on which days. My manager and the desk decide how to correct and inform the report's users. After that I would fix the code, add a test for this case and suggest a daily comparison against a separate source.”

“I enjoy problems where correctness and speed both matter, and markets have both. A price feed never waits, and a trade record must be exactly right. Goldman Sachs advises companies on deals and trades in markets for its clients, so engineers there work close to the business and not far behind it. In college I liked my algorithms and database courses most, and these problems use both. I do not know finance deeply yet, and I see that as something to learn. I would like to begin by understanding how one trade moves from order to settlement.”