85: Reel Friends: Building Social Discovery that Scales to Billions

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

Denne episoden er hentet fra en åpen RSS-feed og er ikke publisert av Podme. Den kan derfor inneholde annonser.

Episoder(91)

84: Trust But Canary: Configuration Safety at Scale

Have you ever wondered how Meta makes config rollouts safe at scale? In this episode, Pascal sits down with Ishwari and Joe to discuss Meta's approach for propagating changes across services in second...

2 Apr 37min

83: Patch Me If You Can: AI Codemods for Secure-by-Default Android Apps

At Meta, even seemingly simple engineering tasks—like updating an API—become monumental undertakings when you're dealing with millions of lines of code and thousands of engineers, especially if the ch...

27 Feb 47min

82: CSS at Scale with StyleX

It's not just Not Invented Here Syndrome. Some technologies like CSS simply don't scale if you're building some of the largest websites on the planet with thousands of engineers committing to the same...

8 Jan 44min

81: From Zero to Polish: Building Meta Ray-Ban Display

You've likely heard of Meta Ray-Ban Display by now — but what's it actually like to work on it? In this episode, Pascal talks to Kenan and Emanuel about the exciting features of Meta's First-Gen Displ...

12 Des 202547min

80: Lowering emissions with the Open Compute Project

In this episode, Pascal talks to Dharmesh J. (DJ) and Lisa about the vision for the open, scalable future of networking hardware for AI and to break down Meta's big announcements from the 2025 Open Co...

14 Nov 202538min

79: Building Android apps in Meta's monorepository with Buck2

How do you keep Android build times under control when your codebase spans tens of thousands of modules and millions of lines of Kotlin? In this episode, Pascal talks with Iveta, Navid, and Joshua fro...

10 Okt 202537min

78: Generating 3D Worlds with AI

Creating 3D assets can be daunting, but does it have to be? Mahima and Rakesh are on a quest to democratize 3D content creation with AssetGen, a foundation model for 3D. They discuss the challenges of...

19 Sep 202536min