10 Optiver Software Engineer (New Grad) Interview Questions (2026)

Outline

STAR. Pick a real optimization — profiling baseline, identifying bottleneck, applying a specific change, measuring improvement. Optiver values measurement discipline. Quantify everything: baseline latency, target, achieved.

Outline

Specific reasons: pure market-maker culture (no clients, no proprietary research bets), flat hierarchy, fast career progression for performers, low-latency engineering challenge, global presence. Avoid 'high pay' as the lead. Reference Optiver's history and engineering specifics if you have read them.

Outline

Bit manipulation. n is a power of 2 iff n > 0 AND n & (n-1) == 0. Reason: powers of 2 have exactly one set bit; subtracting 1 flips all lower bits and clears the set bit. Walk through with examples (4 = 100, 4-1 = 011, AND = 0; vs 6 = 110, 5 = 101, AND = 100 nonzero). O(1) time.

Outline

Horizontal scan: take first string as candidate prefix; for each remaining string, trim candidate until it's a prefix. O(S) total where S is sum of characters. Alternative: vertical scan — character at index i across all strings. Mention divide-and-conquer for theoretical interest. Walk through example.

Outline

Two approaches: (1) Sort both strings, compare. O(n log n) time, O(n) space. (2) Count character frequencies in both, compare maps. O(n) time, O(k) space where k is alphabet size. Walk through; choose hash-based for optimal. Edge cases: different lengths return false, Unicode considerations.

Outline

Reference-counted pointer. Internally: pointer to managed object + pointer to control block (containing reference count and weak count). Copy increments count atomically; destructor decrements. When count reaches 0, deletes object. When weak count also 0, deletes control block. Discuss thread safety: count is atomic, but the pointed-to object is NOT thread-safe automatically. Performance: atomic increment is non-trivial; pass by const& when possible.

Outline

Copy: produces an independent duplicate of an object (deep copy of owned resources). Move: transfers ownership of resources from source to destination, leaving source in a valid but unspecified state. Move is cheaper than copy for resource-owning types (string, vector, unique_ptr). Use std::move to cast to rvalue reference, enabling move constructors. Discuss when move helps (returning large objects, transferring containers).

Outline

OS scheduler picks a new thread, saves current thread's CPU state (registers, instruction pointer, stack pointer) to its task struct, loads new thread's state. If new thread is in a different process, also switch page tables (flushing TLB). Cost: ~1-10 microseconds. Discuss why this matters in trading — context switches in hot paths kill latency. Tools: cgroups, CPU pinning, real-time scheduler (SCHED_FIFO) to reduce switches.

Outline

Flip the coin twice, treat as 2-bit number (00, 01, 10, 11). Map 00→1, 01→2, 10→3, 11→reject and retry. Expected 4/3 trials. Discuss why rejection sampling is needed — direct mapping of 4 outcomes to 3 categories must be non-uniform. Walk through the algorithm.

Outline

Components: (1) Order book: sorted by price-time priority. Buy side max-heap by price; sell side min-heap by price. (2) Match engine: on new order, match against opposite side until exhausted or no overlap. (3) Persistence: append-only event log for recovery. (4) API: receive orders, broadcast trades and book updates. Discuss tradeoffs: in-memory vs persistent, throughput vs latency. Walk through a sample match.