NEW DELHI – The National Payments Corporation of India (NPCI), in collaboration with NVIDIA, has introduced an open reinforcement learning (RL) environment designed for banking artificial intelligence (AI) agents. Unveiled at the Global Fintech Fest (GFF) 2026, this initiative marks a significant step towards advancing AI capabilities within India's financial sector.
Revolutionizing Banking AI Development
Described as the first of its kind for Indian banking, the new environment offers a stable, reusable, and cost-effective platform for developers, banks, fintech companies, and research teams. Its primary purpose is to build, train, and benchmark AI agents using entirely synthetic data. This innovative approach allows AI models to learn from complex, multi-turn banking conversations, with their performance evaluated based on successful outcomes.
A key benefit of using a fully synthetic environment is the ability to conduct extensive training without relying on actual customer data. This significantly strengthens data sovereignty and promotes the responsible development of agentic AI, ensuring customer privacy and security are paramount.
Empowering Conversational AI Assistants
The core objective of this environment is to help conversational AI assistants evolve into sophisticated, tool-using agents capable of resolving a wide range of customer requests within defined operational rules. Financial institutions will now have the means to train diverse AI models on common banking tasks and rigorously benchmark their performance before deployment.
The framework also supports community-driven growth. Participants are encouraged to contribute new tasks, tools, and improvements, fostering an expanding platform that can address additional banking use cases and domains. This collaborative model extends NPCI’s digital public infrastructure approach to AI training, establishing an open and governed foundation for the entire ecosystem.
Built with NVIDIA Technology and Open Standards
The development of this pioneering environment leveraged NVIDIA NeMo’s reinforcement learning framework. It is also designed for contribution to NeMo Gym, an open library dedicated to the evaluation and enhancement of AI agents. Adhering to open standards, including the Model Context Protocol (MCP), the platform facilitates the training and benchmarking of AI models against specific Indian banking tasks.
Developers and institutions will find the environment accessible, allowing them to set up and extend its functionalities for their unique banking needs. By providing a common platform for training, evaluation, and comparison, NPCI and NVIDIA aim to lower barriers to the responsible development of agentic AI and establish a verifiable standard for banking AI agents across India.
NPCI, the umbrella organization behind India’s retail payment and settlement systems, has a proven track record of developing critical digital payment infrastructures such as UPI, RuPay, IMPS, and AePS.