Had an interview with arctic wolf today. I think i have given my worst interview till day.

Mostly because of talking.

First question was designing LFU. I have designed LRU before, didn’t practice it before interview. Its a simple struct with two things: cache of key to item and freq of int(freq) to a linked list of items.

First mess-up was agreing on having to traverse list to get the frequency and update to the newer freq line. And I resorted to using a map to find. It was in the end, I realized the list is a pointer of item.

The mistake was not writing out the Get and engage in a discussion, the cache[key] would give back the item, and item being a doubly linked node, the delete is always O(1)

The worst part was on the system design round. The ask was simple:

Ingestor and Processing pipeline and alerts with AI metrics and summarization.

The FR and Non-FR were produced on a screen. This was generally a drift in-terms of how I have practiced. In here, I would figure out the flow, and list the FR/Non-FR and then try and do the normal design to satisfy the Non-FR.

What went wrong was:

• Things started off with a deviation
  • I should have written down the FR seperately on the board
• I looked the FR and tried to dump everything from memory
• This led me to drawing incomplete structures, loosely connecting ideas, no explanation on the lines drawn
• In General I would go one FR at a time, drilling into failures and non-FR for each component in the FR.
• The interview-er kept prompting to complete the requirements one at a time
• In my head I was trying, to figure out what’s going wrong, rather than recalibrating
• While discussing infra, I missed addressing how to scale servers, rather I completely failed to discuss component failure, because I never drew the right components.

Solutions:

• Practice the already practiced problems from various angles and find a way to get to the flow we know for us.
• Think about the thoughts that are leading to freeze-up and find problems around it
• Shadown practice thinking about interview situation on-line and on-board.

Each of these interviews are a very costly learning experience. The number of accepts is so less, like 1 in 50-60 would probably turn up.