User foundation models for Grab

Summary

Traditional recommendation systems at Grab relied on siloed, manually engineered features that struggled to capture sequential interaction data and cross-service user intent across its superapp ecosystem. To resolve these limitations, Grab developed a custom user foundation model built on a transformer architecture designed to learn from both tabular profile data and time-series clickstream interactions. The architecture unifies diverse data modalities—including categorical identifiers, numerical values, text, and geographical coordinates—by framing inputs as key-value token pairs. Custom positional embeddings and attention masks ensure the model processes tabular attributes as unordered sets while maintaining the sequential order of time-series events. The pre-trained model generates holistic user embeddings and supports direct fine-tuning across downstream tasks such as fraud detection, churn prediction, and advertisement optimization.

Takeaways

• Off-the-shelf LLMs lack domain-specific context for superapp data structures like geohashes, requiring custom architectures capable of handling heterogeneous modalities including geographic coordinates, transaction amounts, and entity IDs.
• Grab standardizes input representations into universal key-value token pairs, pairing column names with attributes for tabular data and event types with entities for sequential clickstream data.
• The transformer architecture uses custom positional embeddings and attention masks to process tabular key-value tokens as an unordered set while preserving the strict chronological sequence of time-series events.

Related reading

• Grab · Jul 16, 2025

  Grab's service mesh evolution: From Consul to Istio

  Grab operated over 1,000 microservices across hybrid infrastructure using Consul alongside a fallback mechanism called Catcher. Single-point-of-failure vulnerabilities in Consul servers and limited support for multi-cluster operations prompted an evaluation of alternative mesh technologies, ultimately leading to the selection of Istio. Grab avoided the standard single-control-plane-per-cluster pattern by deploying multiple external control planes in dedicated Kubernetes clusters arranged in active-active pairs. Migration began in Q4 2024, shifting traffic across AWS and GCP while handling both HTTP and gRPC protocols with gradual traffic-shifting and rollback mechanisms.

  Hilman Kurniawan

• Grab · Oct 21, 2025

  Modernising Grab’s model serving platform with NVIDIA Triton Inference Server

  Grab's machine learning serving platform, Catwalk, experienced mounting technical debt, elevated latency, and rising costs from maintaining disparate inference engines for various frameworks. To address these limitations, the team adopted NVIDIA Triton Inference Server to establish a unified engine, starting with an ONNX migration. They built a custom Triton manager component featuring a proxy layer to translate legacy API requests alongside a server manager that handles model downloads, verification, configuration, and health checks. Within ten days, over half of online deployments migrated seamlessly without requiring client code modifications. The transition reduced p90 latency from 120ms to 20ms on large transformer models and cut average infrastructure spend by approximately 20 percent across evaluated services.

  Daniel Tai

• Grab · Nov 6, 2025

  Grab's Mac Cloud Exit supercharges macOS CI/CD

  Scaling iOS CI/CD workloads across multiple large mobile applications led to substantial expenses and queue delays under rented cloud macOS infrastructure. To mitigate network latency to Git servers and cut compute overhead, Grab relocated its build fleet from a US cloud provider to a colocation data center in Malaysia. The infrastructure deployment relies on over 200 bare-metal Mac minis across 42RU racks, purposefully avoiding virtualization solutions that introduced performance and stability penalties. Phased migration and guardrail monitoring verified cluster stability while achieving 20% to 40% reductions in pipeline and app build times. Operating the regional bare-metal cluster over an estimated three-year hardware replacement lifecycle is projected to yield 2.4 million USD in total cost of ownership savings.

  Madushan Gamage

• Grab · Jun 20, 2025

  Counter Service: How we rewrote it in Rust

  The Integrity Data Platform team rewrote Counter Service, a high-throughput Golang microservice serving event counts for fraud rules and machine learning models, to evaluate the operational return on investment of Rust. Rather than performing a line-by-line translation, engineers approached the service as a black box, reimplementing core application logic from scratch to satisfy established gRPC contracts across Scylla and Redis. The team resolved internal Go tooling dependencies by building custom configuration template parsers using the nom parser combinator and selected targeted open-source crates such as fred.rs and Cadence. Adapting to Rust required navigating cooperative, stackless async execution compared to Go's preemptive concurrency model, alongside managing borrow checker constraints. Ultimately, the rewrite achieved a 70% reduction in infrastructure costs while maintaining comparable service performance.

  Jia Long Loh