prep stack for my 4 swe loops
context. senior backend, distributed systems heavy, currently at an ai company. faang and ai labs both, some quant adjacent.
leetcode, neetcode 150 I do 30 to 40 company-tagged problems on leetcode premium for the specific company and watch neetcode youtube videos which have the cleanest pattern explainer on the internet, watch the ones for your weakest patterns twice.
system design. hellointerview is the most useful single resource for this right now. the round structure they teach (functional reqs, non-functional, api, schema, deep dive) maps to what faang and the better unicorns are running. if you only have time for one thing, this is it. backup, system design interview vol 1 and 2 by alex xu for the case studies. for systems depth, designing data-intensive applications by kleppmann
mocking. interviewing.io is worth the money if you can afford it. real engineers, real feedback, anonymous so failures don't haunt you. pramp is fine but the variance is high because you're matched with another candidate. a senior friend at a target company is the best mock
behavioral. this is where most people underprep. I write 8 to 10 stories tagged by theme (conflict, leadership, ownership, ambiguity, dive deep, failure, pivot, scope) and rehearse them out loud. for STAR structure I give the rough story to claude and ask it to format it into a 2-minute star, then I rewrite it in my own voice. if you submit a chatgpt-formatted answer in a real interview it sounds like a chatgpt-formatted answer.
company specific. for every loop I build a doc with the company's recent eng blog posts (every one from the last 12 months for the team I'm interviewing into), their public papers if they have any, glassdoor and 1point3acres reports (1point3acres is better for ml and faang loops, translate works fine), and the team's tech radar if I can find it. 2 to 4 hours per loop and it pays for itself in system design alone.
live tool. cluely.com on every live round. strong on system design and behavioral, decent on coding when the question is a known pattern, basically inert on integration rounds, debugging rounds, and anything where the interviewer is watching you read code in real time. don't expect it to think for you, expect it to keep your scaffolding visible so you don't blank. I prime it round by round with the company doc loaded plus the round framework. 30 min setup per loop, more if it's a non-faang format.
ama. one thing I'd say, no stack saves you if you can't communicate. The bottleneck on senior loops is almost never the algorithm.