Large product companies running a standardised, centrally calibrated loop: an algorithmic screen, one or two design rounds scaled to level, and a behavioural round scored against published principles. Use this guide to map the interview rounds, scoring signals, common traps, and high-yield questions for FAANG, MAANG, Big Tech roles.
The defining feature is calibration rather than difficulty. Every interviewer works from the same rubric and writes evidence-based feedback, so the panel is trying to place you at a level rather than decide whether you can do the job. That means the coding round is genuinely algorithmic and time-boxed, the design round widens in scope as the level rises, and a strong answer with no stated trade-offs scores worse than a middling answer that names what it gave up.
Treat it as a process gap, not a grudge. Find out where and when decisions are made, get your team's constraints in front of the deciders before the meeting through short written proposals with a comment window both time zones can reach, and escalate a specific costly decision, not the people who made it.
Bitmask DP keeps the cheapest cost for every subset of workers already used. The number of set bits says which task comes next, so there are 2^n states with n moves each, O(2^n * n) time and 2^n memory. At n = 20 that is about a million states, where trying all 20! orders is hopeless.
Plain Dijkstra minimises cost alone, so it can return a cheap route with too many stops, and settling each city once discards a dearer route that still has stops to spare. Make the state city plus edges used, or run Bellman-Ford for exactly K + 1 rounds, each reading only the previous round's distances.
For one grid, number of islands is a flood fill: each unvisited land cell starts a BFS or DFS, O(rows x cols). For land arriving cell by cell, rerunning that is too slow, so use union-find. Each new cell adds one island, and each union with a neighbour in a different set removes one, at near-constant amortised cost.
Evaluate feature flags in an SDK from a cached ruleset, assign users by hashing a per-experiment salt with the user id so assignment is sticky, log exposure where the arms first diverge, gate readouts on an automated SRM check, and make the kill switch a top-precedence rule that works with the control plane down.
A notification system at 50 million a day is sized by bursts and provider limits, not its average of about 580 a second. Accept with an idempotency key, apply preferences once, fan out to per-channel queues with transactional and bulk lanes, rate-limit each provider, and never fail over blindly on a timeout.
Binary search for the rotation point by comparing the middle element with the right end. If it is larger, the break lies to the right; otherwise it is at mid or before it. That is O(log n). With duplicates an equal comparison says nothing, so you drop the right end, and the worst case becomes O(n).
Describe your use of AI tools by task, each with its check: reading cited lines, running code, confirming APIs against the docs for your version. Refuse any answer you cannot check cheaply or that is costly to get wrong, keep restricted data out of unapproved tools, and back it with one real catch.