Uploaded June 2026 | Updated November 2026, 4 days ago
Training data for Meta's recommendation systems was entirely stored in Data Warehouse, structured as relational tables where each row captures labels and snapshotted features at the point of recommendation.
New modeling techniques, such as learning from user sequences and multi-modality, has led to a 10-100x increase in feature size, making the training data increasingly cost-prohibitive due to high duplication. The same user's features are stored repeatedly for every recommendation request, with highly popular content features being duplicated potentially over a million times.
We present a co-designed data and infrastructure in order to address the scaling challenge. By moving features out of training samples into a high-performance indexing storage and implementing model access pattern-aware pushdown optimizations, we have achieved a 10x storage cost reduction for the largest feature: long user sequences.
Want to learn more about the @Scale Conference? Visit our website here: atscaleconference.com
Training data for Meta's recommendation systems was entirely stored in Data Warehouse, structured as relational tables where each row captures labels and snapshotted features at the point of recommendation.
New modeling techniques, such as learning from user sequences and multi-modality, has led to a 10-100x increase in feature size, making the training data increasingly cost-prohibitive due to high duplication. The same user's features are stored repeatedly for every recommendation request, with highly popular content features being duplicated potentially over a million times.
We present a co-designed data and infrastructure in order to address the scaling challenge. By moving features out of training samples into a high-performance indexing storage and implementing model access pattern-aware pushdown optimizations, we have achieved a 10x storage cost reduction for the largest feature: long user sequences.
Want to learn more about the @Scale Conference? Visit our website here: atscaleconference.com
![Live Panel: Behind-the-Scenes with the Team: How Meta Builds Next-Gen Wearable Experiences
Whitney Curry, Jeff Brown and Baiyang Liu, Meta
This was a panel discussion moderated by Whitney Curry, Metas lead for product marketing on wearables, featuring engineering leaders Jeff Brown and Ang Bayan Leo.
[00:11] The conversation explored Metas vision for AI-powered wearable glasses as the next major computing platform, analogous to the evolution from cell phones to smartphones. Jeff and Bayan detailed the significant engineering challenges involved, such as balancing powerful AI with the hardware, power, and form factor constraints of glasses.
[02:05] They discussed the development and technical complexity of specific features, including the Restyle This image creation tool and the on-device Live Speech Translation. The panelists also highlighted the real-world impact and unexpected accessibility benefits of their work, concluding with a look at future challenges related to scaling both the technology and their teams to deliver a step change in hands-free computing.
Jeff Brown was a software engineering manager at Meta who had been with the company for seven years. He supported teams building AI experiences for wearable devices and had previous experience working on camera software, computational photography, and tools for small businesses at Meta.
Ang Leo, who went by Bayan, was a software engineer who had been at Meta for ten years. His work focused on AI glasses, and he had spent his entire tenure building voice assistants for products like Portal, Oculus, and all generations of AI glasses. His recent focus was on AI features designed to break language barriers, such as live speech translation and live captions. Live Panel: Behind-the-Scenes with the Team: How Meta Builds Next-Gen Wearable Experiences](https://i.ytimg.com/vi/_DJ8b7lQvOE/mqdefault.jpg)









