Razorpay, one of India’s largest payments companies, has launched Vulcan, which it describes as India’s first transformer-based AI foundation model for payments, according to medianama.com. The announcement was made in August 2026, marking a significant step in integrating advanced AI into the payments sector.
The company chose to build Vulcan as a foundational model to overcome limitations of traditional machine learning models that are typically unidirectional and use limited signals. Razorpay CEO Harshil Mathur explained that Vulcan creates a relationship map across data, enabling the addition of new use cases such as lending. This approach allows the model to incorporate many small underlying signals that impact outcomes, which conventional ML models often miss.
Vulcan’s development reflects a broader trend in the payments industry toward leveraging AI to enhance decision-making and operational efficiency. By moving beyond single-use ML models, Razorpay aims to reduce noise in predictions and improve accuracy. This positions the company alongside global peers adopting transformer-based models to handle complex financial data and expand service offerings.
Harshil Mathur also highlighted the company’s efforts to mitigate risks such as feedback loops that could affect model outcomes. The launch of Vulcan is part of Razorpay’s ongoing AI initiatives, which include internal AI agents and automation tools. The full interview with Mathur is available on Reasoned, providing deeper insights into the technical and strategic aspects of Vulcan.