Razorpay debuted Vulcan, calling it India's first transformer-based AI foundation model for payments. The company claims Vulcan replaces multiple traditional ML models with a single architecture, tackling payment noise and complex data relationships. This could allow Razorpay to scale new features like lending faster, without retraining entire systems.
How We Got Here
Razorpay CEO Harshil Mathur has been a known early adopter of AI, running a Chief of Staff AI agent and building tools like Openclaw skills for podcasts. The company’s move into foundation models extends this long-standing internal focus on leveraging AI at scale.
The Numbers
- Vulcan processes thousands of signals from billions of historical payments, creating a comprehensive relationship map across data.
- It currently optimizes for four specific outcomes: routing, international fraud, domestic chargebacks, and checkout personalization.
- The model improves with more data flow and allows optimization for new "heads" (outcomes) without requiring a full model retraining.
- Vulcan addresses the challenge of traditional ML models that become noisy with too many signals or struggle to capture granular, unknown insights.
What Happens Next
🇮🇳 Why This Matters for India
For the lakhs of payment-dependent e-commerce startups in Ahmedabad, Chennai, and Jaipur, better fraud detection and success rates from Vulcan could mean 100-200 bps higher net revenue on every transaction.
The Take
Razorpay's Vulcan creates an immediate technical benchmark for all other Indian payments players, especially PhonePe and PayU. They'll now scramble to build or acquire similar foundational AI capabilities to avoid being outmanoeuvred on core metrics like success rates and fraud.
Source:
MediaNama ↗