Goldman Sachs Interview Guide (2026): Engineers and Strats

Goldman Sachs runs on a stack you have almost certainly never touched: a single object database called SecDB that models the firm’s entire risk, and a proprietary scripting language, Slang, that thousands of people write against every day. Goldman has spent years trying to loosen Slang’s grip on that system, first with Java wrappers around SecDB, then with Python access through Marquee and the open-source GS Quant library, and now with individual services being pulled out of SecDB entirely and rebuilt as standalone microservices. (Atlas, if you’ve heard the name, is a different piece of the firm: its low-latency equities trading platform, not a SecDB replacement.) Goldman employs north of 12,000 engineers, more than most of the AI labs, and a large share of them do software in a way no product company does. The interview looks like everyone else’s on the surface. What it is really checking is whether you can do serious engineering inside a bank.

So before anything else, know which door you are walking through, because Goldman splits technical hiring into two tracks and you apply to one.

Engineering and Strats are two different jobs

Software Engineering, the division most people mean by “tech,” builds the platforms: trading and risk systems, the Marquee client APIs, data infrastructure, the internal tooling that ten thousand colleagues depend on. Strategists, universally called strats, sit next to the business. Desk strats price derivatives and manage risk shoulder to shoulder with traders. There are also strats in risk, in core quant modeling, and in engineering-adjacent roles. A strat is a quant developer. You need the math and the code, and you get judged on both.

The loops overlap but the weighting is different. A pure SWE loop leans on data structures, some system design, and a behavioral read. A strat loop keeps the coding and adds probability, expected-value questions, and a math-heavy conversation about statistics or a pricing problem. If you are strong at coding but rusty on brainteasers, apply to engineering. If you like the math and want to sit near a trading desk, strats fits better, and desk-strat comp tends to run higher because it tracks the business more directly.

The process: HackerRank, screens, then a Superday

The front door for most tech roles is an online assessment on HackerRank. Two to three problems, roughly 60 to 90 minutes, auto-graded on hidden test cases. For strats the assessment often bolts on a set of probability and math multiple-choice items, so a coding-only prep plan leaves points on the table. Early-career candidates frequently get a HireVue on top: a recorded video round with a few behavioral prompts and sometimes a light technical question, done on your own time.

Clear the assessment and you get one or two live technical screens with an engineer or strat, usually 45 to 60 minutes in a shared editor, part coding and part walking through something on your resume. The final round is the Superday. It packs four or five interviews into one day, historically in the New York office and now just as often virtual, or run out of Dallas, Bengaluru, or Warsaw, which have grown into engineering hubs in their own right as the firm pushes more headcount outside Manhattan. Expect coding, a design or systems discussion, behavioral questions, sometimes a conversation about how you would operate inside a change-controlled development process, and for strats a probability block. The name is the bank’s, not mine, and it is exactly what it sounds like.

Stage	Format	What it tests	Typical timing
Online assessment	HackerRank, 2-3 coding problems; strats also get probability/math items	Data structures, correctness under a timer	60-90 min, auto-graded
HireVue (early-career)	Recorded one-way video, behavioral plus a light technical prompt	Communication and motivation, “why Goldman”	20-30 min, on your own schedule
Technical screen	1-2 live calls with an engineer or strat in a shared editor	Coding, resume deep-dive, how you reason out loud	45-60 min each
Superday (final round)	4-5 interviews in one day, onsite in NYC, Dallas, Bengaluru, or Warsaw, or virtual	Coding, design discussion, behavioral, process/SDLC questions; probability for strats	Roughly half a day

The coding bar: arrays, hash maps, and finance-flavored prompts

The bar is not FAANG-brutal. Think LeetCode easy to medium: arrays, strings, hash maps, the occasional two-pointer or simple dynamic programming problem, and recurring favorites like sliding windows, monotonic stacks, and “find the next greater element” style questions. The HackerRank grader wants all test cases green, so edge cases and off-by-ones cost you real points in a way a human interviewer might forgive. Get the brute force working first, then optimize if there is time. A correct O(n²) beats a broken O(n).

The phrasing you tend to see:

• “Given an array of daily prices, find the maximum profit from one buy and one sell.” One pass that tracks the lowest price seen so far and the best gain against it; the nested-loop version is correct but the grader’s large cases will time it out.
• “Return whether a string of brackets is balanced.” Push each opener onto a stack, pop and match on each closer, and reject if the stack is non-empty at the end — the unmatched-closer and empty-string cases are where people lose the last few test cases.
• “Count the pairs in an array that sum to a target.” A single pass with a hash map of the values you have already seen takes it from O(n²) to O(n); decide up front whether an element may pair with itself and how to handle duplicates.
• “Given a list of trades, compute the running position and flag when it goes negative.” Keep a running total and test it right after each trade is applied; the points are in the flag rule, so pin down whether “goes negative” means strictly below zero or at-or-below.

That last kind, with a finance flavor bolted onto a standard problem, shows up more in strat and desk-facing loops. The underlying algorithm is still a hash map or a prefix sum. Nobody expects you to know derivatives pricing to answer it.

Probability and brainteasers, mostly for strats

This is the block that separates people who prepped for Goldman specifically from people who ground LeetCode and hoped. Strat interviews reliably include expected-value and probability questions, asked conversationally, where they care about your reasoning as much as the number. A few in the wild:

• “You roll a fair six-sided die. You can bank the roll and get paid that many dollars, or throw it away and roll one more time, with only one reroll allowed. What’s the expected payout if you play optimally?” Solve it backward: if you’re forced into the second roll, it’s worth 3.5 on average, so on the first roll you should keep anything at 4 or higher and reroll a 1, 2, or 3. That gives a 3-in-6 chance of keeping an average of 5, plus a 3-in-6 chance of rerolling into 3.5, for an expected payout of 4.25. It’s the same shape as a market-maker deciding whether to hold a quote or wait for a better one, which is exactly why strat interviewers like it.
• “You roll a die until you get a 6. What is the expected sum of all your rolls?” Break it in two: the number of rolls until a 6 is geometric with mean 6, the last roll is the 6 itself, and the five-or-so rolls before it each average 3, which adds up to 21.
• “A hundred passengers board a plane; the first sits in a random seat and everyone after takes their own seat or a random one if it’s taken. Probability the last passenger gets their assigned seat?” (One half.)
• “Draw two cards from a shuffled deck. Probability both are aces?” It is 4/52 for the first ace times 3/51 for the second, or 1/221; what they are checking is that you condition the second draw on the first ace being gone rather than multiplying two independent 4/52 terms.

The airplane-seat problem and dice-decision problems like the one above are close to canon in these loops. Work through a real quant probability set until conditional expectation and simple recurrences feel automatic. If you freeze on setting up the states, that is the specific weakness to drill.

System design questions built around order flow and market data

Goldman’s design questions skew away from “design Twitter” and toward systems that have to be correct and auditable. Design a service that ingests a market data feed and computes positions in real time. Design an order management system and reason about what happens when a message is dropped or duplicated. Talk about idempotency, ordering guarantees, and how you would reconcile state after a failure, because in a trading system a lost or double-counted message is money and a regulator’s phone call, not a stale timeline.

For senior SWE loops you will also get ordinary distributed-systems material: caching, queues, database choices, consistency trade-offs. The tell that you have thought about this domain is talking about correctness and recovery before you talk about scale. A candidate who opens with “how do we make sure we never lose or replay a fill” is speaking the language of the desk.

SecDB, Slang, and the tools built on top of them

The proprietary stack is the thing recruiters undersell. SecDB, short for Securities Database, is a single object graph that has modeled the firm’s positions and risk for more than three decades, built up by thousands of developers writing Slang against it every day. It is unlike anything you learn in school or at a startup, so your first months are spent learning a system you cannot just Google your way through. The direction of travel is away from hand-written Slang: new work increasingly goes through Java or Python wrappers instead of Slang directly, and individual services are gradually being rebuilt outside SecDB as standalone microservices rather than living inside the object database. On the client-facing side, Marquee exposes the firm’s pricing and risk over web and API, and GS Quant is the open-source Python toolkit built on top of it, worth cloning before an interview so you can speak to it with some fluency.

The real trade-off is that this is a bank. Regulation, change control, and audit trails are real constraints, deploys are more careful than at a product company, and some of your work is plumbing that keeps a regulated business running rather than a shiny feature. In exchange you get problems with actual stakes, unusually strong mentorship in the strat groups, and exposure to how markets and risk work that you cannot get anywhere else. Engineers who thrive there like the depth and do not mind that “move fast and break things” is, for good reason, not the operating model when the thing that breaks is a trading book.

Comp: competitive base, bonus is the wild card

Base salaries are competitive, especially at junior levels, but total compensation trails Big Tech at the senior end, and the reason is structure. Reported new-grad analyst base in NYC tends to land somewhere in the low six figures, with associates higher and VPs higher again. The variable is the bonus, which is discretionary, swings hard with the firm’s year, and at senior levels arrives partly as deferred Goldman stock that pays out gradually over the next several years rather than all at once. That deferral is the core difference from a FAANG offer, where a large slice of comp is liquid RSUs from day one.

So a Goldman VP and a big-tech senior engineer can post similar headline numbers while the cash-versus-deferred mix and the volatility look nothing alike. Do not anchor on a single Blind screenshot. Pull the current ranges from levels.fyi, cross-check Wall Street Oasis for the banking-specific view of base and bonus by title, and in the offer conversation ask directly how the bonus splits between cash and deferred stock and what the vesting schedule looks like. That last question tells you more about the real value than the total figure does.

Where to put your prep hours before Superday

Do timed HackerRank problems, not untimed LeetCode. The interface, the auto-grader, and the pressure of hidden test cases are the actual test, and getting comfortable with that environment is worth more than grinding a hundred hard problems in a nicer editor. Cover arrays, strings, hash maps, two pointers, and basic dynamic programming, and practice getting a working answer out fast rather than chasing the optimal one under a clock.

If you are going for strats, budget at least as much time on probability as on coding. Expected value, conditional probability, and simple recurrences should be reflexes. For system design, read up on how exchanges and order management systems handle ordering and failure, so you can talk about idempotency and reconciliation without hand-waving. And prepare real “why Goldman, why this division” answers, because in a bank behavioral rounds carry more weight than they do at a startup, and “I want to work on hard problems” is not an answer. Know that SecDB, Slang, and Marquee exist and be able to say a sentence about each, and know that Atlas is the equities trading platform, not the Slang replacement, so you don’t trip over it if it comes up. Walking in aware that you are joining a firm with its own language, rather than acting surprised by it, is a quiet signal that you did the homework the rest of the field skipped.