Stating that AI is the third biggest shift for the BFSI sector, Head of Applied AI at Sarvam Vedant Trivedi said the country is still in its first year when it comes to the AI revolution. Yet, the sector is now overwhelmed with hundreds of offerings of voice bots, document AIs, co-pilot licenses resulting in scattered pilots, creating vendor lock-in, fragmented data, and compliance paralysis.
To address this, Sarvam talked of a “core intelligence factory,” a modular stack separating fast-moving, commoditised layers like GPU infrastructure and models from layers where it is crucial for institutions to retain control and IP like governance, context engineering, orchestration, and learning loops.
“We have taken inspiration from the core banking system built over the last two, three decades. They have the strategic autonomy to plug and play different layers of the stack as they wish and move at the pace of frontier,” said Trivedi.
Different layers
The bottom layer relates to infrastructure, which Sarvam advises sourcing from a secure, centralised “token factory” inside the country, considering the growth in chipset development and other hardware. The models layer helps institutions keep from committing to a single provider. The governance layer deals with model risk management and a centralised “kill switch” to enforce policy compliance across AI agents at runtime. The context engine layer determines how a bank’s data is pre-packaged and represented to agents for accuracy and cost-efficiency. The final orchestration layer improves performance.This can help banks move their technologies in step with AI innovation without being tied down to a single hyperscaler.
Published on September 11, 2026