Katkat

Company

Mercor interview questions

44 practice questions in the style of Mercor's interviews, covering Behavioral & Leadership, ML System Design, Software Engineering Fundamentals, Coding & Algorithms, System Design, Machine Learning, Statistics & Math.

Questions44

3 easy38 medium3 hard

Your progress0/44

Mercor interview questions by the numbers

Last updated

Katkat has 44 Mercor interview questions. The largest topics are Software Engineering Fundamentals (43%), System Design (20%) and Coding & Algorithms (9%). By difficulty: 7% easy, 86% medium and 7% hard. The most common round is Technical Screen (77%). Frequent concepts include distributed-systems, system-design, code-review and combinatorics.

Difficulty

  • Easy7% · 3
  • Medium86% · 38
  • Hard7% · 3

Interview rounds

  • Technical Screen77% · 34
  • Onsite23% · 10

Roles

  • Software Engineer73% · 32
  • Machine Learning Engineer27% · 12

Frequent concepts

  • distributed-systems9% · 4
  • system-design7% · 3
  • code-review5% · 2
  • combinatorics5% · 2
  • idempotency5% · 2
  • order-statistics5% · 2

Showing 31–44 of 44 results

  • Design and implement an in-memory referral network as a directed acyclic graph (DAG) supporting cycle-safe edge insertion, depth-bounded reachability queries, and expected reward computation under geometric decay.

    MercorMachine Learning EngineerCoding & AlgorithmsOnsite
  • A payment endpoint calls an external provider to charge the customer and then writes a record to its own database. The question asks what can go wrong and how to redesign the flow, testing failure analysis for crashes and timeouts, safe retries, durable payment state, recovery through status checks, webhooks and reconciliation, and lock placement.

    MercorSoftware EngineerSystem DesignTechnical Screen
  • A memoization cache built on a plain Python dictionary wraps an expensive function and is now called from many threads at once. The question asks what goes wrong and how to fix it, testing race-condition analysis, lock scope and its throughput cost, duplicate-work prevention, and exception handling for waiting callers.

    MercorSoftware EngineerSoftware Engineering FundamentalsTechnical Screen
  • A probability brainteaser: two drinks are ordered at each of two coffee shops and every drink receives a quality score, and you must find the probability that both drinks from the first shop outscore both drinks from the second. It tests symmetry arguments, counting rankings, and stating independence and tie assumptions.

    MercorSoftware EngineerStatistics & MathTechnical Screen
  • A probability puzzle in which two customers each buy two cups of coffee of random quality, and you find the chance that one customer's cups are better than the other's. It tests clarifying what better means, using symmetry over rankings, checking the result a second way, and reasoning about ties and correlated cups.

    MercorSoftware EngineerStatistics & MathTechnical Screen
  • A startup background round: introduce yourself, walk through a recent project probed from a product angle (who owns each decision, what done means, success metrics), explain what the company does and why you want to join, and describe your side project. It also tests how you use AI tools for on-call automation, code review, and design discussions.

    MercorSoftware EngineerBehavioral & LeadershipTechnical Screen
  • Calculate conditional comparison probabilities for coffee quality scores under continuous and discrete distributions, accounting for ties and sample permutations.

    MercorSoftware EngineerStatistics & MathTechnical Screen
  • Assess and debug a compact decoder-only Transformer implementation. Candidates trace tensor dimensions through causal multi-head self-attention, identify flaws in next-token loss alignment and KV-cached generation, and establish equivalence between cached and uncached forward passes.

    MercorMachine Learning EngineerMachine LearningOnsite
  • A code review question on a short Python function that turns a table of rows into CSV text. Candidates must find its correctness and performance problems and write a fixed version, which tests delimiter handling, string building in Python, non-string cells, CSV quoting and escaping rules, and how to verify the fix.

    MercorSoftware EngineerSoftware Engineering FundamentalsTechnical Screen
  • Tests understanding of asynchronous race conditions, memory leaks, and eviction in single-threaded event loops versus multithreaded synchronization in languages like C++.

    MercorSoftware EngineerSoftware Engineering FundamentalsTechnical Screen
  • This question tests your ability to apply Breadth-First Search (BFS) on an augmented state space where path state includes both coordinates and remaining resources.

    MercorSoftware EngineerCoding & AlgorithmsTechnical Screen
  • A verbal sorting-theory discussion: explain how to sort faster than quadratic insertion sort, identify quicksort's worst case, count the non-empty subsequences and the permutations of an n-element array, and design a correct sort that is deliberately slower than O(n^2). It tests asymptotic analysis, recurrence reasoning, and basic combinatorics.

    MercorSoftware EngineerSoftware Engineering FundamentalsTechnical Screen
  • Explain how to add a Stripe payment integration to a backend service: how to handle errors and retries without charging a customer twice, what payment data to store, and which database to use given ACID and availability needs. Tests idempotency, failure handling, and data modeling for payments.

    MercorSoftware EngineerSoftware Engineering FundamentalsTechnical Screen
  • A verbal algorithms question: name the worst sorting algorithm that still always sorts correctly, analyze its running time, and explain how you would compute every permutation of the input exactly once. It tests reasoning about factorial complexity and knowledge of permutation enumeration methods.

    MercorSoftware EngineerSoftware Engineering FundamentalsTechnical Screen

Mercor interview experiences

All 22 →
  • No-Coding Technical Screen with Code-Reading Questions, Did Not PassMercor · Software Engineer · Technical Screen · Oct 2026
    Rejected

    I applied online. There is no recruiter screen - first call is a HM screen where they ask you basic questions. Next screen is a 30 minute technical screen with no coding. Mine had 4 questions - Tell me the worst sorting algorithm - (n!) and explain how you would calculate each permutation You visit two coffeeshops A and B and order 2 drinks from each. Each drink gets a quality score. What is the…

  • Technical Screen with Bloom Filter and Payment Idempotency DebuggingMercor · Software Engineer · Technical Screen · Oct 2026

    Almost all of these were questions I had already seen in interview experience posts! Worst sorting algorithm: I talked about enumerating every permutation and checking whether each one is sorted. Coffee shop: A and B each buy two cups of coffee. What is the probability that A's two cups taste better than B's two cups? There are 1M user_ids. How do you efficiently check whether an id exists, and a…

  • 30-Minute Rapid-Fire Technical Screen, Rejected After Three QuestionsMercor · Software Engineer · Technical Screen · Sep 2026
    Rejected

    The first round at Mercor was a 30-minute live technical interview. It was rapid-fire Q&A, where you try to answer as many questions as possible. I only got through 3 questions, so I didn't pass. The first question was to explain in detail a sorting algorithm faster than O(n^2). I gave merge sort. The second question was to give one slower than O(n^2). For brute force you can shuffle and then com…

  • A Rushed 30-Minute Screen on Sorting, Probability, and a Cache Race ConditionMercor · Software Engineer · Technical Screen · Sep 2026

    The interviewer was a principal from Caltech. He talked fast and used a lot of technical jargon, like a college professor would. It was rushed, only 30 minutes. Explain a sorting algorithm better than O(N^2). Merge sort — recursively split the array into left and right halves until there are only 2 elements left, then merge by inserting the smaller of the two into a new array and returning it (N…

  • Today I had a tech screen for Mercor's Applied AI role. The debugging question: The other rapid-fire questions: What is idempotency, and why do we need idempotent operations? What is RAG, and how is it related to fine-tuning? How would you handle context limit exceeded in an LLM application? LLM calls are usually stateless. How would you make an LLM application stateful? How would you handle prom…

More companies