You've probably spotted those little circles of your friends' faces popping up on Facebook Reels. They look simple enough, but building them was a proper engineering challenge. In this episode, Pascal chats to Joseph and Subasree about Friend Bubbles, a feature that surfaces which of your close friends have been watching and reacting to the same Reels as you.

We get into the details of how prefetching keeps things snappy without wrecking scroll performance, why the team's ML model had to move from survey-based friend rankings to real-time interaction signals, and the surprising discovery that showing fewer bubbles actually made the whole feature click. If you've ever underestimated a "simple" feature, this one's for you.

Got feedback? Send it to us on Threads (https://threads.net/@metatechpod), Instagram (https://instagram.com/metatechpod) and don't forget to follow our host Pascal (https://mastodon.social/@passy, https://threads.net/@passy_). Fancy working with us? Check out https://www.metacareers.com/.

Links

Timestamps

Intro 0:06
Meet the Engineers: Backgrounds and Roles 1:53
Goals and Aspirations in Video Recommendations 4:20
The Origin of Friend Bubbles 4:41
Defining Success: Metrics and User Experience 5:40
Client-Side Constraints and Challenges 6:57
Feature Description: What Are Friend Bubbles? 8:31
Initial Challenges and Performance Issues 9:29
Architectural Changes for Performance 11:34
Impact of Performance on User Experience 15:14
Addressing Client-Side Challenges 16:58
Model Development: From Surveys to Interactions 20:07
Evolving the Model: Real-Time Data and User Interactions 23:35
Exploring Model Training and Performance 24:58
Feedback Loops and User Engagement 25:56
The Role of AI in Development 29:49
Collaboration Across Teams 32:17
Future Directions for Friend Bubbles 34:02
Safe Rollout Strategies for Features 35:22
Outro 37:31
Bloopers 38:27