[Flash talk] Fraud detection at scale using AI @ Razorpay
summary
Razorpay captures billions of events, extracts 200+ parameters into 300–400 features, and claims real-time fraud decisions within 200–300 ms. Features include user journey timers, IP/phone/location fingerprinting and third‑party city/tier/payment‑capability data. Models combine incremental learning classification on labeled history and LSTM sequence models for behavior, deployed inside a Flink app with Kafka ingestion and POSTed scores and dashboards for merchant action.
AI-written summary · about
related talks
35mAI Powered Fraud Detection for Payments & Transactions
16mAI-Driven Behavioral Analytics for Real-Time Fraud Prevention
26mAI-Powered Fraud Detection in Cloud-Native Digital Payments
13mTransforming Financial Services with Real-Time Analytics
13mBehavioral Analytics for Real-Time Fraud Prevention
18mReal-time Fraud Detection using Deep Learning
08mAI-Powered Fraud Detection in Financial System
37mThe Role of Machine Learning in Today’s Payment Ecosystems