Low-Latency C++ & Systems

Optiver's software-engineering rounds turn on how the machine really works. The interviewers say so in their own words: the technical interviews are "not standard LeetCode puzzles or brainteasers." They test command of one object-oriented language, how data structures sit in memory, and computer architecture, networking, concurrency, and memory management. This is the low-latency layer, and it is the round that pure algorithm preparation walks straight past.

The Optiver angle

At a market maker, microseconds are money, so the bar is mechanical sympathy and determinism, not algorithm range. Optiver quotes two-sided prices across global markets from its own capital, and the edge lives in the path from market data in to an order out. Its engineering writing describes collapsing the layers between those two points and running native code end to end. The interview follows the work: expect questions on cache behavior, allocation, lock-free structures, and the network stack, defended out loud. A candidate who can profile and reason about the machine reads as someone who has shipped fast code, not someone who has only read about it.

One question runs under every section here, asked many ways: where does the time go, and is it predictable? Latency that is low on average but spikes without warning is often worse here than latency that is slightly higher but flat, because a trading system lives and dies on the tail.