Razorpay Launches Vulcan, AI Foundation Model Built for Payments
Vulcan is not an LLM such as ChatGPT. Razorpay describes it as a payments-focused foundation model trained to identify patterns in how money moves through its network.
Bengaluru-based fintech company Razorpay has launched Vulcan, a transformer-based AI foundation model built specifically for payments, as it looks to improve transaction success, fraud detection, payment routing and checkout personalisation.
Razorpay said Vulcan has been trained on nearly 4 billion payments and 3 trillion data points, analysing around 3,000 signals per transaction. The model was developed using technology and infrastructure from NVIDIA and AWS.
Unlike general-purpose AI models, Vulcan is designed to understand payment behaviour across merchants, payment instruments, issuing banks and payment gateways. Razorpay aims to use a common AI layer for functions such as routing, fraud detection, risk assessment and checkout personalisation.
Early components of Vulcan have already been tested on live transactions, with customers including Blinkit, Bachatt and redBus using some of the capabilities.
Razorpay said early deployments have delivered an 8–10% improvement in payment success rates. The company also reported that the system detected eight times more international card fraud and identified five times more fraudulent or disputed transactions without increasing the number of alerts.
On its Magic Checkout platform, Razorpay said Vulcan helped 40% more shoppers see their preferred UPI app, contributing to an additional 1–2 lakh purchases per month.
The model can analyse transaction signals to determine the payment route most likely to succeed, identify fraud patterns, assess risks associated with cash-on-delivery orders and recommend suitable payment methods during checkout.
Razorpay CEO and co-founder Harshil Mathur said the model is designed to improve as it processes more payment data. Vulcan is not an LLM such as ChatGPT. Razorpay describes it as a payments-focused foundation model trained to identify patterns in how money moves through its network.
The company said NVIDIA GPUs were used for training and inference, while AWS and Amazon SageMaker supported the model’s development, training and deployment.
Razorpay plans to expand Vulcan across additional payment functions, including authentication, routing, fraud detection and lending, as digital payments continue to grow in India.