A while back I interviewed for an E6 INFRA role at Meta. Every single interview felt positive, but I ultimately got cut at the hiring committee round, with no specific feedback. Pretty frustrated, but nothing to do except move on.
Phone screens: I got "Er Liu" and "Odd Mutant" — aced both.
Onsite coding: "Commander" and "Yao San San" — also aced both, and the follow-up edge cases the interviewer threw at me afterward were easy to work through too.
AI coding: I drew the "maximum subset" problem. I passed the first two test cases, but the last two wouldn't finish running. For this problem, passing the first two counts as a pass — I've never seen anyone actually finish the third. I tried every backtracking optimization I know and it still didn't work. Maybe DP has some hope, but this problem doesn't ask for the length of the max subset — it wants you to print out the actual max subset — so DP isn't really a fit; memory would blow up.
SD1 — Online Auction: All you really needed to hit was the pub/sub pattern, SSE for real-time bid-result notifications, and a message queue partitioned by auctionId — once I covered those key points, the interviewer didn't push back much. I improvised a bit on the rest. QPS estimation also ate up some time.
SD2 — Location Based Service Search: This one was a simplified version — input is lat/long plus a radius, output is the top-K locations within that range. Note the problem never specifies what kind of location this is — it could be a fixed place like a Yelp business, or a moving one like nearby Uber drivers. The focus is on how you search. I gave three options: 1. Elasticsearch, 2. Postgres with the PostGIS extension, 3. a custom-built geospatial index. For the first two, the interviewer just nodded without much reaction, signaling they already knew them, so I focused on how to build a custom spatial index for the search — the principles behind geohash and quadtree, the tradeoffs between searching static vs. dynamic locations, the search patterns and specific algorithms, quadtree memory estimation, and the database sharding choice (a hybrid of geohash and quadtree). I talked for about 20 minutes on this, and the interviewer seemed genuinely satisfied, nodding along in approval the whole time.
BQ was a bunch of standard questions, asked quite a few, I don't remember most of them — the whole conversation ran about an hour.
After I finished everything, the recruiter's feedback was very positive — they replied to every follow-up email instantly, emojis flying, telling me not to worry. I thought I had it in the bag. Then it went to the hiring committee and got rejected. The recruiter sent one templated rejection email and never responded to another message after that, no specific reason given. The INFRA role has a one-year freeze on reapplying, but I was told I could still apply to other roles — no freeze there.
Discussion
Loading comments…