Juspay Interview Questions and Answers (2026)

Store for each node its parent, a locked-by user id, and a count of locked descendants. lock checks that the node is unlocked, that its locked-descendant count is zero, and walks up the parents to check no ancestor is locked; on success it sets the owner and increments the count on every ancestor. That is O(height) per call instead of scanning the subtree. unlock checks the owner, clears it and decrements the ancestors’ counts. upgrade checks that no ancestor is locked and the count is positive, collects locked descendants with a DFS, verifies they all belong to the user, unlocks them and locks the node. Build a map from node name to node at input time so lookups are O(1).

A single global lock is correct but serializes every call. A finer design locks the path a call touches: acquire per-node mutexes from the root down to the target node in a fixed top-down order, which prevents deadlock because every thread acquires locks in the same order. Check and update the ancestor counts while holding those locks, then release bottom-up. Another approach is optimistic: read the path without locks, then use atomic compare-and-set on each node’s state and retry if anything changed. Explain the trade-off: path locking is simpler to prove correct, while compare-and-set scales better under low contention. Write a test that runs thousands of concurrent lock calls on overlapping subtrees and checks that no ancestor and descendant are ever locked together.

Because every node has at most one exit, each component contains at most one cycle. Walk from every unvisited node, recording the step number at which each node was reached during this walk. If you reach a node already seen in the current walk, you have found a cycle: sum the indices from that node around the loop. If you reach a node visited in an earlier walk, stop. Mark everything visited so the total work is O(N). Return the maximum cycle sum, or -1 if no cycle exists.

From each starting cell, follow the single outgoing edges and record the distance to every cell reached, stopping when you hit -1 or revisit a cell. Then for every cell reached from both starts, compute the sum of the two distances and return the cell with the smallest sum, breaking ties by the smallest index or as the problem states. Each walk is O(N), so the whole solution is O(N).

If the target must appear as a subsequence, check greedily whether “Juspay” is a subsequence of the input by scanning once with a pointer into the target; if it is, the answer is the input length minus six, otherwise it is impossible. If the question asks for a contiguous result, you need a substring equal to the target, so search for it and remove everything else. Clarify which version is meant and whether matching is case-sensitive before coding.

A deadlock happens when threads wait on each other in a cycle and none can proceed. It needs four conditions together: mutual exclusion, hold and wait, no preemption and circular wait. The simplest prevention is to remove circular wait by always acquiring locks in a global order, which is exactly what makes top-down locking safe in Tree of Space. Other options are acquiring all locks at once, using tryLock with a timeout and backing off, or avoiding shared mutable state.

Strong static types catch whole classes of bugs at compile time, such as passing an amount in the wrong currency if currencies are separate types. Pure functions make business logic easier to test and reason about, and immutable data avoids many concurrency bugs. Juspay’s open-source EulerHS framework for building web backends is written in Haskell. You do not need to know Haskell for campus interviews in the reports reviewed here, but this is a reasonable answer if an interviewer asks.

For campus SDE hiring: an online coding test with three tree and graph problems, a hackathon Part A on HackerEarth built around the Tree of Space problem, a Part B that extends the same code to be thread-safe and faster over a full day, a technical interview on DSA, OS, DBMS, networks and OOP, and an HR round. Some drives add an aptitude or MCQ round and an elimination round between the hackathon parts.

Graph and tree problems of medium to hard level. The online test commonly uses the Converging Maze set: Largest Sum Cycle, Maximum Weight Node and Nearest Meeting Cell, in 90 minutes on platforms such as Otomeyt or D2C. Other reported questions include a binary search problem, the minimum characters to remove to turn a string into Juspay, and a tree problem.

The hackathon Part A problem: implement lock, unlock and upgrade operations on an N-ary tree, where a node can be locked only if none of its ancestors or descendants is locked. Candidates solve it on HackerEarth in about three hours, then in Part B make the solution thread-safe and reduce its time complexity while explaining the approach.

A long session, reported as eight hours or a full day on Slack, where candidates optimize their Part A code, make it thread-safe for concurrent lock and unlock calls, and explain their reasoning to evaluators. A 2025-batch candidate said mentors were available but asking too many questions could hurt the evaluation.

A 2025-batch campus SDE candidate reported a CTC of 27 LPA. Juspay does not publish campus packages, so treat this as one reported figure and confirm the fixed and variable split on the offer letter.

Not in the campus interview reports reviewed here, which cover DSA, concurrency and CS fundamentals. Juspay does use Haskell: its open-source EulerHS framework for web backends is written in Haskell, and one recruitment guide says some backend teams may ask about functional programming. Knowing why pure functions and immutability help with concurrency is useful background.

DSA plus CS fundamentals. Reports list computer networks, operating systems, DBMS, OOP and problem-solving; a 2025 candidate had graphs, arrays, binary search, BFS and DFS, OS and socket programming; and a candidate was asked how Gmail works on the backend.

Juspay is headquartered in Bengaluru, India, and lists hubs in Bangalore, Sao Paulo, San Francisco, Dublin and Singapore. Campus and fresher hiring is for Bengaluru, and some interviews have been held at other Karnataka locations such as Mysore.

Reports range from about a day for compressed campus drives to several months, because the online test, hackathon parts and interviews are often spread over different dates. One 2025-batch candidate got the hackathon invite on the same day as an HR meeting about it.

Yes. Some drives assign candidates to SDE or React developer tracks based on the first test, and one 2025 report describes a Learnathon with Postman API testing challenges and test-case design for automated refunds instead of the thread-safety hackathon.