Meta Software Engineer Interview Questions

I would first think about flattening the matrix and doing binary search, but that is only valid if the matrix ordering supports it. If each row and column is sorted, I would start from the top-right corner and move left or down depending on whether the current value is too large or too small. The reason I choose that approach is that it eliminates one row or one column at each step, giving O(m + n) time. If the matrix is globally sorted in row-major order, then I could use binary search for O(log(mn)) time. I would also mention edge cases like empty matrices and single-row or single-column inputs.

Before designing it, I would clarify the file sizes, supported formats, latency expectations, and whether thumbnails need to be available immediately. My design would separate upload, storage, and processing into different stages so the user does not have to wait for thumbnail generation to finish. I would store the original image in object storage, send a processing event to a queue, and let a worker service generate thumbnails asynchronously. The reason I choose this design is that image processing can be expensive, and decoupling it keeps the upload path fast and reliable. I would also think about retries, corrupted files, and content validation.

I would use two pointers, one for each string. I would scan through the larger string and advance the pointer in the smaller string whenever the characters match. If I reach the end of the smaller string, then it is a subsequence. The reason I choose this approach is that it is simple, O(n) time, and O(1) space. If there are many queries against the same large string, I would mention preprocessing it for faster repeated lookups.

I would first clarify whether comments need threading, sorting by popularity, moderation, or real-time updates. At a high level, I would store posts and comments separately, with comments linked to a post ID and optionally a parent comment ID for threads. I would also think about pagination, because large comment trees can be expensive to load all at once. The reason I would keep the read model flexible is that comment ranking and nesting often evolve over time. I would also mention caching, abuse prevention, and eventual consistency if comment counts are updated asynchronously.

I would first think about a brute-force search of all possible combinations, but that explodes quickly. To optimize it, I would use dynamic programming and build up the minimum number of coins needed for each amount from zero to the target. The reason I choose DP is that the problem has overlapping subproblems and optimal substructure. The time complexity is O(amount times number of coin types), and the space complexity is O(amount). I would also mention edge cases like unreachable amounts, zero amount, and duplicate coin denominations.