NPCI announces domain-specific AI language Model FiMI built for India’s payments ecosystem

New Delhi: National Payments Corporation of India (NPCI) on Tuesday unveiled FiMI (Finance Model for India), a domain-specific language model built for India’s payments ecosystem, at the India–AI Impact Summit 2026.

Developed in-house, FiMI currently powers NPCI’s UPI Help Assistant and is aimed at addressing gaps seen in general-purpose large language models when deployed in high-scale, high-trust payment workflows.

FiMI has been designed to understand the complexities of Indian payment systems, including UPI, covering areas such as transaction dispute resolution, mandate lifecycle management, and regulatory queries.

The model underwent continuous pre-training and fine-tuning on Indian financial and synthetically generated payments data, enabling structured reasoning, tool invocation, and multilingual support.

NPCI said this payments-native approach allows the system to function reliably in high-volume, real-world environments where accuracy and consistency are critical.

The model is already live nationwide through NPCI’s UPI Help Assistant, an AI-driven conversational support platform for UPI users.

Built on an agentic AI framework, the assistant supports multi-step reasoning to address payment queries, grievance redressal and mandate-related requests.

It currently operates in English, Hindi, Telugu and Bengali, with more Indian languages planned over the next six to eight months.

Following its production deployment, NPCI has published a detailed technical paper on arXiv outlining FiMI’s data curation, training methods and evaluation metrics.

The move, NPCI said, underscores its commitment to transparent and responsible AI development for financial infrastructure.

Chief Technology Officer Vishal Kanvaty said FiMI reflects NPCI’s push to build purpose-driven financial AI while strengthening trust and collaboration across the ecosystem.

NPCI is also exploring advanced architectures, including Mixture-of-Experts, to enhance scalability and

efficiency.