Uploaded November 2025 | Updated September 2026, 2 weeks ago
At Ray Summit 2025, Nick Buehrer and Chongyu Zhou from Grab share how the company is building a powerful user embedding foundation model to understand the full, multi-modal journeys of users across Southeast Asia’s leading super app.
They explain how a single Grab user interacts across a wide range of services—ride-hailing, food delivery, payments, and more—creating a complex, interconnected behavioral footprint. To learn from this rich tapestry of signals, Grab set out to develop a unified model capable of capturing a holistic view of each user.
The result is a set of general-purpose user embeddings: numerical representations that serve as a universal feature set for numerous downstream applications. These embeddings eliminate the need for siloed, hand-engineered features and unlock more personalized, consistent, and scalable ML capabilities across the platform.
If you’re interested in multimodal modeling, foundation models for personalization, or large-scale production ML systems, this session offers a deep look at Grab’s approach.
Liked this video? Check out other Ray Summit breakout session recordings youtube.com/playlist?list=PLzTswPQNepXllnU0C36WtkC0dqkAoDulh
Subscribe to our YouTube channel to stay up-to-date on the future of AI! youtube.com/c/anyscale
🔗 Connect with us:
LinkedIn: linkedin.com/company/joinanyscale
X: https://x.com/anyscalecompute
Website: anyscale.com
At Ray Summit 2025, Nick Buehrer and Chongyu Zhou from Grab share how the company is building a powerful user embedding foundation model to understand the full, multi-modal journeys of users across Southeast Asia’s leading super app.
They explain how a single Grab user interacts across a wide range of services—ride-hailing, food delivery, payments, and more—creating a complex, interconnected behavioral footprint. To learn from this rich tapestry of signals, Grab set out to develop a unified model capable of capturing a holistic view of each user.
The result is a set of general-purpose user embeddings: numerical representations that serve as a universal feature set for numerous downstream applications. These embeddings eliminate the need for siloed, hand-engineered features and unlock more personalized, consistent, and scalable ML capabilities across the platform.
If you’re interested in multimodal modeling, foundation models for personalization, or large-scale production ML systems, this session offers a deep look at Grab’s approach.
Liked this video? Check out other Ray Summit breakout session recordings youtube.com/playlist?list=PLzTswPQNepXllnU0C36WtkC0dqkAoDulh
Subscribe to our YouTube channel to stay up-to-date on the future of AI! youtube.com/c/anyscale
🔗 Connect with us:
LinkedIn: linkedin.com/company/joinanyscale
X: https://x.com/anyscalecompute
Website: anyscale.com
