Meta Engineers Reveal the Hidden Complexity Behind Friend Bubbles for Reels

Friend Bubbles Feature Relies on Extensive Machine Learning Overhaul

Meta's new Friend Bubbles feature on Facebook Reels—highlighting Reels your friends have watched and reacted to—appears deceptively simple. However, the engineering team behind it has revealed that the feature required a complete rethinking of machine learning models and cross-platform optimization, handling billions of interactions daily.

Meta Engineers Reveal the Hidden Complexity Behind Friend Bubbles for Reels — Source: engineering.fb.com

In a newly released episode of the Meta Tech Podcast, software engineers Subasree and Joseph detailed the journey from concept to deployment. “What looks like a simple UI bubbles actually represents a system that must understand social signals, ranking, and personalization at an unprecedented scale,” Subasree said.

The Machine Learning Evolution

To power Friend Bubbles, the team evolved the underlying ML model to balance friend content with general Reels recommendations. “Initial models struggled to weigh friendship proximity without sacrificing relevance,” Joseph explained. The solution involved training separate model architectures—combining collaborative filtering with graph-based social connections—to create a hybrid system.

This model now processes hundreds of millions of friend interactions per day to surface the most engaging Reels.

iOS vs. Android: A Surprising Discovery

One unexpected finding was how differently iOS and Android users interacted with bubbles. “Android users tapped frequently, while iOS users tended to scroll past,” Joseph noted. This forced the team to optimize UI behavior per platform, adjusting bubble size, animation, and placement.

The engineers also revealed a “surprising discovery” that made the feature click: combining reaction types with watch time provided a stronger social signal than either metric alone. This insight allowed real-time updates to reflect not just who watched, but how they engaged.

Background

Friend Bubbles launched earlier this year as part of Meta’s push to increase social discovery on Facebook Reels. The feature shows a circular avatar of a friend who recently interacted with a Reel, along with a small reaction icon. Tapping the bubble takes the user directly to that Reel.

It builds on previous social indicators but is the first time Meta has connected direct friend activity so prominently within the Reels feed. The feature now serves billions of recommendations daily, requiring infrastructure that can handle massive real-time updates.

What This Means

This engineering approach signals Meta’s commitment to blending algorithmic content with social signals. Successful features like Friend Bubbles could lead to deeper integration of friend-related data across other Meta products, including Instagram Reels and Facebook Watch.

For users, it means a more personalized and socially-aware experience where recommendations are influenced not just by what you like, but by what your friends actually watch and engage with. The full podcast episode is available on Spotify, Apple Podcasts, and Pocket Casts. Meta also encourages feedback via Instagram, Threads, or X.

For more on Meta’s engineering challenges, see the Background section.

Tags:

Meta Engineers Reveal the Hidden Complexity Behind Friend Bubbles for Reels

Friend Bubbles Feature Relies on Extensive Machine Learning Overhaul

The Machine Learning Evolution

iOS vs. Android: A Surprising Discovery

Background

What This Means

Related Articles

Recommended

Discover More