NPCI Deploys AI to Enhance UPI Infrastructure
The National Payments Corporation of India (NPCI) has launched its in-house AI model, FiMI, to improve the UPI ecosystem. In parallel, NPCI is leveraging a partnership with Nvidia to integrate AI across its operations for enhanced payment security and fraud detection. The initiative aims to increase the reliability and efficiency of UPI as transaction volumes continue to grow.
- The FiMI (Finance Model for India) is a specialized small language model trained in-house by NPCI on Indian financial and synthetic payments data to better understand the nuances of the local payments ecosystem, including transaction disputes and regulatory queries. - A key application of the FiMI model is powering the UPI Help Assistant, a pilot initiative for AI-driven conversational support that provides multi-step reasoning to assist with grievance redressal and is currently available in English, Hindi, Telugu, and Bengali. - The collaboration with Nvidia provides NPCI with access to advanced AI computing platforms and the Nemotron family of open models to build a sovereign, payments-native AI foundation that aligns with India's data sovereignty regulations. - With UPI processing over 16-21 billion transactions monthly, the AI integration aims to enable real-time, dynamic risk scoring and fraud pattern detection by analyzing transaction metadata, behavioral deviations, and device signatures without slowing down transaction speeds. - This AI push addresses the challenges of scaling the UPI infrastructure, which has faced issues like technical glitches, outages, and interoperability problems as transaction volumes have grown. - In 2025, the UPI platform processed 228 billion transactions worth a cumulative ₹300 trillion, marking a 33% increase in volume and a 21% increase in value compared to the previous year. - NPCI is also exploring federated learning models, where participating banks can train a shared fraud detection algorithm locally on their own data, enhancing system-wide intelligence while preserving data privacy. - The long-term goal is to move from use-case-specific AI tools to a foundational, scalable AI layer for the entire payments ecosystem, supporting everything from operational intelligence to multilingual voice-based payments.
