Citadel Software Engineer Interview: Coding & Quant Round

Before you prep a single problem, settle one thing: which Citadel is on your calendar. The name sits on two separate companies under Ken Griffin, and they run their own loops. Citadel is the hedge fund, the buy side, where engineers build the platform, data, and research tooling that the investment process runs on. Citadel Securities is the market maker, the sell side, quoting roughly a quarter of US cash equities and about a third of listed options. There the engineering is low-latency C++ threaded through the order path. Same brand, same brutal bar, different center of gravity in the interview.

The fund's software loops read like a faster, harder version of a big-tech screen. Strong algorithms, language-agnostic, graded hard on speed and correctness. The market maker keeps that bar and stacks performance and optimization on top. And for anything that touches trading or pricing, it adds the probability and mental-math round that is the quant fingerprint of this firm. One disambiguation is worth doing once. None of this is The Citadel, the military college in South Carolina. Its Honor and Duty creed has nothing to do with the trading floor, and only surfaces because search engines confuse the two.

The Citadel angle

Half the value of preparing for Citadel is routing yourself to the right track before you start drilling. A core backend or data-infra role at the fund is coding-dominant and may never show you a brainteaser. A quant developer, quant researcher, trading-adjacent engineer, or the early-career NXT track almost always gets probability and market-making games on top of the algorithms. Confirm the entity and the role family with your recruiter, then weight your practice: fund SWE leans language-agnostic algorithms, Securities leans C++ and latency, and the quant tracks add the probability canon. Guessing wrong here wastes the scarcest thing you have, which is prep time before a short-fuse loop.

How the assessment runs

The gate for most software and campus pipelines is an online assessment, usually on HackerRank, and it is where the funnel narrows hardest. The common shape is two problems in roughly sixty-five to seventy minutes, auto-graded against hidden test cases, skewed medium-hard with the occasional hard. You typically get a ten-day window to open the link once it lands, but the clock inside is tight. The Citadel Securities C++ pipeline often looks different: one problem in about seventy minutes, framed as an optimization or refactor rather than a blank page, and sometimes served on CodeSignal instead. Live rounds run in a shared editor, Coderpad-style. Some show you only a single visible test, so you cannot lean on the grader to tell you when you are done. The dedicated Online Assessment lesson breaks down the platform, the clock, and how it is graded.

• Platform: HackerRank OA, live Coderpad. The assessment is usually HackerRank, sometimes CodeSignal for Citadel Securities. Live coding runs in a shared editor, occasionally with just one visible test case, so you verify correctness yourself.
• OA shape: 2 problems, ~65 to 70 min. The software norm. Auto-graded on hidden tests, medium-hard leaning hard. Citadel Securities C++ pipelines often run one ~70-min optimization problem instead. Formats drift by year, region, and entity.
• Full pass expected: Partial credit rarely advances. The hidden cases target the edge you skipped: empty input, a single element, self-loops, overflow, mod arithmetic. A timeout on a slow solution is a first-class failure, not a near miss.
• Feedback: Fast or silent, not fixed. Early stages can turn around next morning; experienced loops report eight-plus business days or a silent reject. Offers can be short-fuse. Do not read slow feedback as a no, and do not assume speed.

The Citadel angle

A correct, working solution to a hard problem can still be a reject, so the bar here is optimality and the right approach, not "it runs." Candidates report solving a shortest-path variant correctly with depth-first search and passing the visible case, then getting cut. The intended answer was breadth-first search, and the interviewer wanted the shortest-path instinct recognized before any code was written. Another was rejected on a median-from-a-stream problem whose only named miss was a missing base case for the empty stream. The lesson is margin: reach for the optimal complexity the first time, finish with time to spare, and surface the edge cases yourself instead of waiting to be told. On this loop, passing tests is the floor, not the verdict.

The families that come up

Citadel does not hand out a question list, and it is one of the more opaque firms on this axis. But the shape of what gets asked is steady across candidate reports, and it is lopsided toward a few families. The tilt is more contest-flavored than a clean big-tech screen. Counting problems want a closed form or a recurrence mod 1e9+7. Constraints get tuned so the naive solution times out. And the occasional puzzle rewards math over a memorized template.

Dynamic programming and counting-with-modulo sit on top, which is the clearest way Citadel diverges from a typical FAANG screen. Prefix-sum and hashing arrays, graph and shortest-path work, and heap-and-stream problems fill the middle. Bit and number-theory problems lean Securities-flavored, and a Trie or autocomplete design turns up in phone screens. The C++ optimization refactor takes correct-but-slow code and drives it to a single linear pass. That is the market-maker signature, so it earns its own treatment later in the section rather than a slice of a chart.

The Citadel angle

The probability, brainteaser, and mental-math round is the quant gate, and treating it as central to every Citadel software loop is the most common way people misprepare. For quant researcher, quant developer, trading-adjacent, and NXT candidates it is a real round. Expect expected-value and fair-price games, combinatorics, Bayesian updating, and gambler's-ruin and random-walk. Then comes make-me-a-market pricing, where the interviewer trades against your quote and watches whether you can be picked off. It often opens with a timed arithmetic sprint, many questions on a short clock, because pricing fast under pressure is the job. But pure backend and infra loops at the fund are coding-dominant and may skip it entirely. The dedicated probability lesson in this section teaches the canon and the commit-to-a-price method; route to it only if your track calls for it.

What this section covers

Five lessons follow. The first four carry the algorithm families in roughly the order they matter. Each opens with how the firm frames the pattern and the variants it favors. Then it hands you the most-asked problems as runnable exercises, with worked follow-ups stacked the way the round stacks them. The fifth is the quant round, taught as method rather than runnable code. Where a lesson rests on a base algorithm, the full walkthrough lives in the Foundations course and is linked, so these pages stay on what is specifically Citadel.

Common questions