TechPrufer
Get started
Grab
Grab logoGrab · Feature trail

Fraud Risk Rule Engine

1 post · Oct 2019 – May 2021

What this trail covers

Grab's fraud detection and prevention capabilities are enhanced by Griffin, an in-house anti-fraud risk rule engine. Archivist, a new distributed tracing, statistics, and feedback system, complements Griffin by providing a centralized, performant, and flexible platform for logging, storing, and querying ML/rule-based prediction events. Archivist supports user-defined entities and custom fields, a shared dedicated service for event processing, and a user portal for querying, prediction stats, and

All Grab trails
  1. Building a Hyper Self-Service, Distributed Tracing and Feedback System for Rule & Machine Learning (ML) PredictionsIntroduced Archivist, a new distributed tracing, statistics, and feedback system for ML/rule-based predictions. Archivist addresses limitations of previous logging methods (plain text logs, lack of fine-grained access control, manual schema updates) by providing a centralized, performant, and flexible solution. Key features include user-defined entities and custom fields, a shared event processing service with Kafka and Elasticsearch, data lake integration with S3 and PrestoDB for data mining, and a user portal for querying, statistics, and feedback with role-based access control. This system reduces development time for new business scenarios and improves investigation efficiency.
  2. Griffin, an Anti-fraud Risk Rule Engine Making Billions of Predictions DailyThis post details the evolution of Grab's fraud detection system from a manual, code-based rule implementation to an in-house rule engine called Griffin. Griffin allows data scientists and analysts to self-service rule creation and deployment via a web portal, significantly improving efficiency and accuracy. The system handles billions of predictions daily with low latency, supporting complex, dynamic datasets and offering features like scenarios, actions, segments, and dynamic values. The architecture separates data orchestration from rule-based prediction, and Python was chosen for its expressive power, with optimizations to ensure low latency.