Meta

Software Engineer

The Meta technical screen had a strong AI focus. The first part was a deep dive into an AI system I had personally designed, with the interviewer asking what problem it solved and how I approached the system.

Anthropic

Software Engineer (Performance)

The Anthropic Performance / Infrastructure SWE process moved through a recruiter screen, hiring-manager conversation, and technical coding phone screen before the onsite invitation. There was no OA in this process.

LinkedIn

Software Engineer

The LinkedIn technical phone screen contained two coding questions.

Citadel

Software Engineer

The Citadel online assessment was a performance-engineering problem rather than a standard LeetCode-style implementation question. I was given an existing C++ implementation of a function named root_node. The code was logically correct, but far too slow for the required production-level performance target. The task was to refactor the implementation without changing its behavior and bring execution time below a strict 100-microsecond limit on large inputs. The problem tested algorithmic complexity, C++ performance awareness, mathematical simplification, and the ability to optimize existing code while preserving exact semantics.

Arize AI

Software Engineer

The Arize technical screen focused on a multithreading problem. The task was to coordinate three dedicated thread classes so that the numbers from 0 through n were printed in strictly increasing order, with each number handled by the thread responsible for that category. The main challenge was not the numeric logic itself, but synchronization: ZeroThread, OddThread, and EvenThread had separate responsibilities while still needing to produce one globally ordered output.

Datadog

Applied ML Scientist

The Datadog LLM Applied Scientist technical screen contained two roughly 25-minute design questions. The first focused on performing Ordinary Least Squares regression when a high-dimensional dataset is too large to fit in memory, while the second asked for a real-time monitoring and paging system for detecting degraded server performance. I felt reasonably good about the technical discussion, but the feedback afterward was that I spent too much time writing out my design in the shared notepad instead of maintaining an active discussion with the interviewer. The interviewer also appeared to expect deeper exploration of the design choices.