LessonThis is a lesson restated in new words. Any figure in it is an example from the source, not this desk's record.
Three early interviews, at Citadel, Akuna, and AQR, produced a filter this lecturer still gives students. Apply for the seat you actually want. Notice what the person across the table cares about. And do the work of that seat before anyone is paying you to do it. The combination, after years of study and ordinary failed rounds, is what this account credits for a run of offers.
Why it matters
Technical interviews mix partial differential equations, combinatorics, and optimization with a human who has to sit near you. Students often hedge: interview as an engineer while hoping to trade, or the reverse. This account says that hedge shows. The Citadel round was a quantitative developer seat. The interviewer had a particle-physics doctorate and had moved over after a postdoc. The conversation opened on backgrounds, then on pricing: a parabolic PDE and the link to a risk-neutral conditional expectation.
Combinatorics came next, permutations and combinations. Treating the prompts as too easy, and reaching for Python's ready-made combinatorics tools, did not land as a joke. The functions had to be written out, including the difference between returning a value and yielding one in a recursive call. The round then moved to memory allocation and latency. The questions were recognizable. The answers were not at hand, and, more telling, the curiosity that usually shows up for a stuck math problem did not show up here.
The idea in plain words
The first lesson was role fit. A developer interview exposed a lack of interest in being a developer, and the interviewer could see it. The advice is to stop hedging. If the seat you picture is trading, interview for trading. Do not collect engineering rounds as a backup story while you would rather be on a desk.
That lesson was not absorbed on the spot. Akuna and AQR were already booked, and the focus stayed on getting through those days. Akuna was a trading interview. Probability, statistics, and mental math were straightforward and were answered, yet they barely moved the interviewer. The energy changed when the conversation turned to trading the candidate had already built: mid- and high-touch systems in Java, Python, and a little C++. Entry, exit, how the market was being thought about, and what "edge" meant were the questions that mattered.
The lesson there is attention. People hire people. One interviewer wants a hard programming puzzle or a stochastic integral. Another cares about judgment, because the two of you may share a row of desks. When you notice genuine interest, stay on that subject and show you care about it. That is rapport, and it is also evidence you want the job in front of you.
AQR was a research interview. It opened on convex versus non-convex optimization and on deriving least squares with matrices, which is calculus in matrix form if you have seen it. The discussion then widened to numerical methods, getting stuck at a local extremum, gradient descent, and stochastic gradient descent, the family most modern routines belong to. The questions became esoteric. The candidate did not know every algorithm named, but wanted the answers, and spent that evening on Google Scholar and Stack Overflow. This was before chatbots.
How to use it
Put the three notes together. Target the seat you would accept. In the room, track what draws the interviewer in, and meet that interest with substance. Before the room, immerse: build, research, or trade with no firm and no paycheck, because you would do it anyway. The lecture's word for that is passion, defined as fulfilling the role before money is attached. After that immersion, odd questions feel less foreign, and you have opinions about methods instead of only definitions.
What usually goes wrong
Hedging across developer, trader, and researcher rounds advertises indifference. So does performing tricks the interviewer did not ask for, such as importing a library when the point was to derive the count. Freezing on a topic you dislike is information. Freezing on a topic you love, then never following up, wastes the AQR-style opening.
Questions people ask
What failed in the developer round?
Interest. Memory and latency questions were understood as questions, but the answers were not known and were not wanted. The seat itself was the mismatch.
What changed the trading interview?
Correct mental math did less than a concrete discussion of systems already built, including how entries, exits, and edge were defined.
How did this person handle rejected rounds?
Not by assuming it would all work out. By treating the search as the only route home, then going back to algorithms that had been unfamiliar the same day.