NPCI’s FiMI AI model to handle UPI disputes

It uses an agentic AI framework for multi-step reasoning tasks.

The National Payments Corporation of India (NPCI) released the Finance Model for India (FiMI)—a domain-specific language model serving the Indian payments ecosystem—to address limitations in general-purpose large language models for payment workflows.

Launched at the India-AI Impact Summit 2026 on 17 February, the model powers NPCI’s UPI Help Assistant, handling transaction disputes, mandate lifecycle management, and regulatory queries, according to a press release.

It uses an agentic artificial intelligence framework for multi-step reasoning to address payment queries and grievance redressal, and supports English, Hindi, Telugu, and Bengali, with plans to add more Indian languages within six to eight months.

FiMI underwent pre-training and fine-tuning on Indian financial and synthetic payments data, targeting accuracy and consistency in high-volume transaction environments, the press release said.

NPCI published a technical paper on the arXiv platform detailing data curation, training methodology, and evaluation results. The organisation is researching model architectures, including Mixture-of-Experts, for scalability and domain intelligence.