Bloomberg Coding Interview: What Gets Asked

Bloomberg's coding rounds are built from recognisable interview problems, most of them medium difficulty, with very little of the puzzle-box character some finance-adjacent companies go in for. What makes this section worth reading is the distribution. Bloomberg concentrates on a different subset of the standard syllabus than its neighbours do, and preparing against the generic distribution wastes a large fraction of your time.

Across Bloomberg's loops, two categories sit far above where they sit anywhere else, and one sits well below.

What actually gets asked

Grid traversal leads the outright frequency table: a Premium question counting is the single most-asked problem at Bloomberg. But it is design-in-code that changes how you should prepare, because it recurs across the whole loop. There is no single round to clear and be done with.

The Bloomberg angle

At most companies, "design a data structure" is a single problem you might get once. At Bloomberg it runs through the whole loop. Candidates report a journey tracker that records taps in and out and averages travel times between stations. They also report a least-recently-used cache, a stack that reports its minimum in constant time, and a structure with constant-time insert, delete and random selection. Browser history and a leaderboard with score updates and rank queries come up too. That cluster is the second, fourth, and several more of the most-asked problems here, which is not a distribution you will find at a Google or a Meta. Look at the product and it makes sense. The Terminal is thousands of small stateful services answering queries about market data under latency constraints that customers notice. Interviewers ask you to build small correct stateful things because that is what the job is. Treat these as the highest-yield problems in your preparation, not as a curiosity.

The linked-structure weighting

The second departure is linked structures. Flattening a multilevel doubly linked list is close to a Bloomberg signature: it appears here at a rate that barely registers in other companies' corpora. Copying a list with random pointers, adding two numbers represented as digit lists, and reversal in its several variants all recur steadily. Doubly-linked structures specifically are weighted heavily, which is partly an artefact of how often the cache problem comes up.

The corresponding good news is dynamic programming. Word Break is essentially the only DP problem that recurs, with its harder variant appearing occasionally. Multi-dimensional DP, the kind that dominates preparation for some other companies, is not where this loop concentrates. You can deprioritise it with a clear conscience.

Two problems, and what that does to you

A Bloomberg technical round usually carries two problems. Candidate accounts are consistent about this across phone screens and onsite rounds alike. After introductions you have roughly fifty working minutes for both.

Reskins, and follow-ups that interrogate your choice

Problems here arrive dressed differently, and the reskin is part of the assessment. A Premium question counting shows up as clouds and clear sky rather than land and water. A Premium variant asks for every pair that reaches the target and asks you to print them rather than return one. A grid problem gains a follow-up midway through where diagonals now count as connected, and your code is expected to absorb that without a rewrite.

The follow-ups are the more distinctive half. They tend not to ask for more code. They ask why you chose what you chose. After a first-unique-character problem, candidates report being asked why a fixed-size character array is the better structure than a hash map here. After the constant-time insert-delete-random structure, they are asked to explain precisely what makes each operation constant time and what hashing is doing underneath to get you there.

The lessons that follow are ordered by weight. Design-in-code comes first because it is both the highest-frequency category and the one most people under-prepare. Linked structures follow for the same reason. Grids, arrays and intervals come next as the conventional core, and stacks and parsing last.