The round of abnormal Ai is pretty sh*t and abnormal. They ask you to read a codebase which I skimmed through in like 10 mins and claude (allowed). Then they (some non-india south asian interviewer) again mentioned you still have 3 mins to go through code. I was like - let me understand the question first then I'll deep-dive into specific component to lear more. Then he gave me a question after I started acting on it. He said sorry I gave the wrong question and it was still a vague question.
The repo did was - processed all security events like collection/ingestion, ranking, threat levelling based on some rules, created alerts, created apis on top of it and put it to database.
-
(This was the wrong question given first) We've to allow users to suppress some rules (it can be complex rules like based on geo-ip (existing in code) and other rules).
-
The enchrichment layer currently hardcodes based on some threats (geo-ip, history, 1 more) clients want more configurablibility without touching platform code. Implement a plugin based mechanism to ensure no code touching by clients. (On these lines, very vague)
After 20 mins of struggling and explaning them how can we add decorator pattern and clients can choose which threats they want to use/avoid etc I resorted to claude to code it completely for me.
I'll mostly get a No. I mean I've given others interviews where a code base is involved but atleast the ask/question is clear as to what is to be achieved/implemented finally. This was a HLD cum LLD plus coding round.