[Meta] ML Research Scientist Interview Experience (New "Coding with AI" Round included) - Detailed Timeline, Questions & Mistakes (Rejected)

First off, I want to say thank you to this community. I am writing this to give back, as I read so many posts here during my preparation that helped me navigate the process. I didn't get the offer, bu

First off, I want to say thank you to this community. I am writing this to give back, as I read so many posts here during my preparation that helped me navigate the process. I didn't get the offer, but I hope my experience, especially the mistakes I made, can help someone else land the job. Here is a full breakdown of my loop for the

ML Research Scientist role, including the new "Coding with AI" round.

Timeline *

Nov 19: First contact from the Recruiter. *

Dec 12: Finished the Online Assessment (OA). *

Dec 18: Recruiter confirmed I

passed the OA. * (Originally scheduled for early Jan, but I had a conference and asked for a rescheduling). *

Jan 18:

Round 1 Behavioral. *

Jan 21:

Round 2 Coding (Standard). *

Jan 26:

Round 3 Coding with AI. *

Jan 28:

Round 4 ML System Design. *

Feb 11: Rejection Notice.

Stage 1: Online Assessment (OA) *

Duration: 1 hour. *

Format: 4 Problem-Solving questions. Two easy questions, one medium and one hard. Forgot the exact questions. *

Performance: I solved 3 questions with the optimal solution. For the 4th question, I used a Brute Force approach. *

**

Passed.**

Stage 2: The Full Loop

**

Round 1 Behavioral (45 mins)** Standard questions about work style, past challenges, and conflict resolution. *

My Critical Mistake: I was doing really well until one specific question. The interviewer asked a question, and I started my answer by saying, "I'll be honest answering this..." * The interviewer smiled and said, "Yes, being honest is good." * I replied, "Not always, since sometimes you have to be prepared for fancier answers." *

Reflection: Looking back, this was likely a major red flag. I tried to be too "real" and it came off as unprofessional or manipulative. Lesson learned: Choose your words.

**

Round 2 Standard Coding (45 mins)** *

Preparation: I solved the first 60 Meta-tagged questions (Free on Taro) and about 100 questions from the NeetCode 150 (started with these mid 2025). *

Question 1: A variation of LC 543 (Diameter of Binary Tree). * The Twist: Instead of counting edges, the requirement was to count nodes. * The Mistake: I was overconfident. I looked at the problem, thought "I've seen this," and wrote the full solution in 5 minutes. However, because I didn't notice it was a variation, my code failed. I wasted 15 minutes debugging perfectly good code for the wrong problem before realizing the difference. The change was to add left + right + 1. * Figured out time and space complexity correctly. *

Question 2:

LC 32 (Longest Valid Parentheses). * Performance: I had not solved this one before. I managed to discuss the optimal solution and implement about 80% of the logic, but time ran out. The interviewer stopped me before I could finish the implementation. * Figured out time and space complexity correctly.

**

Round 3 Coding with AI (1 Hour - New Format)** This is Meta’s new interview style where you are given a full software project (4 or 5 files, including test cases) and an AI agent to help you. *

The Task: You need to fix bugs in the project to pass the test cases. *

My Experience: I actually performed very well here. There were bugs distributed across different files. I fixed the issues and

passed 11 out of 12 test cases. * When I got to the last test case, the interviewer stopped me and said, "You already did so well, no need to fix the last one." *

Constraint: Twice I tried to query the AI, and he stopped me, saying, "You need to give it a closer look yourself." He wanted to verify my manual debugging skills. I ended up not using the AI at all. *

Resource/Gatekeeping: I know many of you are looking for the specific question details for this round. *

Note: I prepared using free resources because I couldn't afford the "Coding with Minmer" subscription. They have a few free questions discussed in their channel but mostly the video are members-only access. * The Deal: I will share the exact problem details and project structure in an update to this post ONLY IF Coding with Minmer agrees to provide a 25% discount code (Promo: Gaza25) for 20 people. I want to make sure others who are struggling financially can access the prep material I couldn't.

**

Round 4 ML System Design (45 mins)** *

Question: Design a Places-to-Visit Recommendation System (similar to Google Maps) that includes a category filter, nearby locations, user's past interactions, etc. *

Preparation: I fully read and summarized "ML System Design Interview" by Ali Aminian and Alex Xu, as well as Hello Interview videos and blogs. *

Performance: I am not an expert in ML theory (I have decent experience but lack deep theoretical knowledge), but I stuck to the structure from the Ali & Alex book. I kept talking, discussing trade-offs, and drawing the components exactly as the book recommends. I'm unsure if this was enough, but I felt I followed the "meta" (pun intended) for this round.

Summary of Mistakes to Avoid 1.

Read the Question Carefully. 2.

Watch Your Words: Don't be so comfortable answering questions. 3.

Don't Rely on the AI: In the AI round, be prepared for the interviewer to disable the tool to test your raw skills.

The Outcome Usually, after the loop, you get one of three responses: 1.

Team Match: You did well and move to finding a team (Offer likely). 2.

Follow-up: You did well generally but missed a spot, so they schedule one more interview. 3.

Rejection: This is what I got on Feb 11th. Good luck to everyone else currently interviewing!