NPCI, NVIDIA unveil open AI training environment for banking agents using synthetic data; check details
NPCI said the framework provides a stable, reusable and cost-effective platform for institutions to develop and evaluate banking AI agents
- Sep 9, 2026,
- Updated Sep 9, 2026 5:07 PM IST
The National Payments Corporation of India (NPCI) has announced an open reinforcement learning (RL) environment for banking artificial intelligence (AI) agents, developed in collaboration with NVIDIA, at the Global Fintech Fest (GFF) 2026.
Described as the first initiative of its kind for Indian banking, the environment is designed to enable developers, banks, fintech companies and research teams to build, train and benchmark AI agents using synthetic data. The initiative allows AI models to learn from multi-turn banking conversations, with their performance evaluated based on outcomes.
READ THIS: IAMAI signs 3-year MoU with Saudi Arabia to strengthen fintech, startup ecosystems
AI agents to handle banking requests
The environment is aimed at helping conversational AI assistants evolve into tool-using agents that can resolve customer requests within defined rules. NPCI said the framework provides a stable, reusable and cost-effective platform for institutions to develop and evaluate banking AI agents.
Banks and fintechs will be able to train different AI models on common banking tasks and benchmark their performance before deploying them. The use of a fully synthetic environment means training can take place without using real customer data, supporting data sovereignty and responsible development of agentic AI.
Built with NVIDIA technology
The environment has been developed using NVIDIA NeMo’s reinforcement learning framework and is designed for contribution to NeMo Gym, an open library focused on the evaluation and improvement of AI agents.
It follows open standards, including the Model Context Protocol (MCP), allowing institutions to train and benchmark AI models against Indian banking tasks. Developers and institutions will also be able to access, set up and extend the environment for additional banking tasks and domains.
ALSO READ: 'India's fintech story has been amazing... A big champion of this is UPI': PM Modi
Open framework for banking AI ecosystem
NPCI said the environment has been designed to support community-driven growth. Participants can contribute new tasks, tools and improvements, enabling the platform to expand into additional banking use cases.
The initiative also extends NPCI’s digital public infrastructure approach to AI training by creating what it described as an open and governed foundation for the wider ecosystem. By providing a common platform for training, evaluation and comparison, NPCI said the collaboration aims to lower barriers to the responsible development of agentic AI and establish a shared, verifiable standard for banking AI agents in India.
NPCI, the umbrella organisation responsible for operating India’s retail payment and settlement systems, has previously developed major digital payment infrastructure including UPI, RuPay, NACH, IMPS, NETC, AePS and e-RUPI.
The National Payments Corporation of India (NPCI) has announced an open reinforcement learning (RL) environment for banking artificial intelligence (AI) agents, developed in collaboration with NVIDIA, at the Global Fintech Fest (GFF) 2026.
Described as the first initiative of its kind for Indian banking, the environment is designed to enable developers, banks, fintech companies and research teams to build, train and benchmark AI agents using synthetic data. The initiative allows AI models to learn from multi-turn banking conversations, with their performance evaluated based on outcomes.
READ THIS: IAMAI signs 3-year MoU with Saudi Arabia to strengthen fintech, startup ecosystems
AI agents to handle banking requests
The environment is aimed at helping conversational AI assistants evolve into tool-using agents that can resolve customer requests within defined rules. NPCI said the framework provides a stable, reusable and cost-effective platform for institutions to develop and evaluate banking AI agents.
Banks and fintechs will be able to train different AI models on common banking tasks and benchmark their performance before deploying them. The use of a fully synthetic environment means training can take place without using real customer data, supporting data sovereignty and responsible development of agentic AI.
Built with NVIDIA technology
The environment has been developed using NVIDIA NeMo’s reinforcement learning framework and is designed for contribution to NeMo Gym, an open library focused on the evaluation and improvement of AI agents.
It follows open standards, including the Model Context Protocol (MCP), allowing institutions to train and benchmark AI models against Indian banking tasks. Developers and institutions will also be able to access, set up and extend the environment for additional banking tasks and domains.
ALSO READ: 'India's fintech story has been amazing... A big champion of this is UPI': PM Modi
Open framework for banking AI ecosystem
NPCI said the environment has been designed to support community-driven growth. Participants can contribute new tasks, tools and improvements, enabling the platform to expand into additional banking use cases.
The initiative also extends NPCI’s digital public infrastructure approach to AI training by creating what it described as an open and governed foundation for the wider ecosystem. By providing a common platform for training, evaluation and comparison, NPCI said the collaboration aims to lower barriers to the responsible development of agentic AI and establish a shared, verifiable standard for banking AI agents in India.
NPCI, the umbrella organisation responsible for operating India’s retail payment and settlement systems, has previously developed major digital payment infrastructure including UPI, RuPay, NACH, IMPS, NETC, AePS and e-RUPI.