In this episode, we have a real mock interview for a Staff Machine Learning (E6) role at Meta, where the candidate is asked to design the recommendation system behind Instagram Reels. This means choosing which short videos to show to billions of users in real time, based on their behavior and interests.

The candidate has a strong grasp of ML fundamentals and proposes modern architecture choices like multitask learning and multi-stage ranking. However, they ultimately do not pass the interview—mainly due to time management and not addressing key practical concerns like feedback loops, feature freshness, and production-readiness. The interviewer offers detailed, actionable feedback that gives you a clear picture of what sets apart a good answer from one that meets the E6 bar.

If you’re preparing for ML system design interviews at Meta, Google, or other top-tier tech companies, this interview is full of insights to help you sharpen your strategy, improve your pacing, and avoid common pitfalls.

Sign up to book coaching or to watch more interviews in our showcase: https://www.interviewing.io

See the interviewer’s feedback and transcript here: https://start.interviewing.io/showcase/Mek40HIliiP0

Or view other Meta interviews: https://interviewing.io/mocks?languag=&company=meta

Disclaimer: All interviews are shared with explicit permission from the interviewer and the interviewee, and all interviews are anonymous. interviewing.io has the sole right to distribute this content.

⸻

Timestamps:

00:00 Introduction and interview setup

00:53 Problem presented: Instagram Reels recommendation system

17:00 Candidate defines ML framing and objective

38:00 Discussion of candidate generation and ranking model

45:00 Interview ends and candidate self-assessment

46:30 Feedback begins: time management, pacing issues

50:00 Why this would not pass the E6 bar

56:00 What the candidate did well and what was missing

1:03:00 Final takeaways from the interviewer