Meta E4 SWE-ML Menlo Park
Summary
I interviewed for an E4 SWE-ML position at Meta in Menlo Park, which included a phone screen and a full loop with coding, behavioral, and ML system design rounds. Despite my efforts, I received a rejection email just 15 minutes after my last interview, with vague feedback about strengthening my experience.
Full Experience
My interview process for the E4 SWE-ML role at Meta in Menlo Park began with a phone screen, which consisted of two coding questions completed in 45 minutes. The full loop then followed, comprising two coding rounds, one behavioral round, and one ML System Design round.
During the coding rounds, I encountered 2D matrix problems, specifically variations of the 'Number of Islands' question, a 'Merge sorted intervals' problem, and questions involving 'Binary Tree Maximum Path Sum', and a 'Balanced Parentheses' variation. I focused on techniques like two-pointers, stacks, and queues for BFS. I found the interviewers to be quite rigid, insisting on implementation on their screen, which made it challenging to articulate my optimal approaches, as they didn't always align with their preferred solution methods. This felt somewhat frustrating, especially compared to a smoother experience I had with a mid-west startup.
The behavioral round involved 6-7 questions, which I answered using the STAR method. I felt I had to repeat myself multiple times, as the interviewer, despite having a DS background, seemed to lack managerial experience. My standard questions about product requirements and optimization metrics revealed that the team might not have had particularly interesting work, focusing primarily on writing well-articulated PSCs every six months.
For the ML System Design round, I was asked to design Instagram ranking. I presented a standard High-Level Design (HLD) after spending about 5 minutes on problem formulation and assumptions. I elaborated on features, user and video metadata embeddings, and explained the necessity of offline processing. I proposed a two-tower model for relevance and ranking, and discussed user engagement metrics like NDCG, Recall, and precision at K. Unfortunately, I struggled to fully explain model deployment. Throughout this round, I felt the interviewer showed little interest, forcing me to reiterate my explanations for feature sets and metrics.
Overall, my impression was that the interviewers were unprofessional. I received a rejection email within 15 minutes of my last interview, with the recruiter stating they couldn't provide any feedback and suggested I 'make my experience stronger by taking some classes / work (somewhere better).'
Interview Questions (5)
I was given a 2D matrix problem, specifically focusing on the 'Number of Islands' problem and its variations.
I encountered a coding question involving merging sorted intervals.
One of the coding questions involved finding the maximum path sum in a binary tree.
I was asked a coding problem that was a variation of the balanced parentheses problem.
I was tasked with designing the Instagram ranking system. I started with a standard High-Level Design (HLD) after 5 minutes of problem formulation and assumptions. I then focused on features, user and video metadata embeddings, and explained why we need it to be offline. I proposed a two-tower model for relevance and ranking, and discussed user engagement metrics like NDCG, Recall and precision at K. I could not explain model deployment effectively.