Monetary providers has lengthy been on the forefront of adopting technological improvements. Right now, generative AI and agentic programs are redefining the trade, from buyer interactions to enterprise operations.
Prem Natarajan, govt vp, chief scientist and head of AI at Capital One, joined the NVIDIA AI Podcast to debate how his group is constructing proprietary AI programs that ship worth to over 100 million clients.
“AI is at its finest when it transfers cognitive burden from the human to the system,” Natarajan mentioned. “It permits the human to have that rather more enjoyable and expertise that magic.”
Capital One’s technique facilities on a “check, iterate, refine” strategy that balances innovation with rigorous threat administration. The corporate’s first agentic AI deployment is a chat concierge that helps clients navigate the car-buying course of, reminiscent of by scheduling check drives.
Relatively than merely integrating third-party options, Capital One builds proprietary AI applied sciences that faucet into its huge information repositories.
“Your information benefit is your AI benefit,” Natarajan emphasised. “Proprietary information permits you to construct proprietary AI that gives enduring differentiated providers in your clients.”
Capital One’s AI structure combines open-weight basis fashions with deep customizations utilizing proprietary information. This strategy, Natarajan defined, helps the creation of specialised fashions that excel at monetary providers duties and combine into multi-agent workflows that may take actions.
Natarajan burdened that accountable AI is key to Capital One’s design course of. His groups take a “duty by way of design” strategy, implementing sturdy guardrails — each technological and human-in-the-loop — to make sure protected deployment.
The idea of an AI manufacturing unit — the place uncooked information is processed and refined to supply actionable intelligence — aligns naturally with Capital One’s cloud-native expertise stack. AI factories incorporate all of the parts required for monetary establishments to generate intelligence, combining {hardware}, software program, networking and improvement instruments for AI functions in monetary providers.
Time Stamps
1:10 – Natarajan’s background and journey to Capital One.
4:50 – Capital One’s strategy to generative AI and agentic programs.
15:56 – Challenges in implementing accountable AI in monetary providers.
28:46 – AI factories and Capital One’s cloud-native benefit.
You May Additionally Like…
NVIDIA’s Jacob Liberman on Bringing Agentic AI to Enterprises
Agentic AI allows builders to create clever multi-agent programs that cause, act and execute advanced duties with a level of autonomy. Jacob Liberman, director of product administration at NVIDIA, explains how agentic AI bridges the hole between highly effective AI fashions and sensible enterprise functions.
Telenor opened Norway’s first AI manufacturing unit in November 2024, enabling organizations to course of delicate information securely on Norwegian soil whereas prioritizing environmental duty. Telenor’s Chief Innovation Officer and Head of the AI Manufacturing unit Kaaren Hilsen discusses the AI manufacturing unit’s fast improvement, going from idea to actuality in beneath a 12 months.
Imbue CEO Kanjun Qiu on Reworking AI Brokers Into Private Collaborators
Kanjun Qiu, CEO of Imbue, explores the rising period the place people can create and use their very own AI brokers. Drawing a parallel to the PC revolution of the late Seventies and ‘80s, Qiu discusses how fashionable AI programs are evolving to work collaboratively with customers, enhancing their capabilities somewhat than simply automating duties.
