How Capital One Delivers Multi-Agent Systems with Rashmi Shetty - #765

Key Takeaways

• •

  Multi-agent systems solve complex goal-oriented tasks

  “We moved from a classic ML world to a world where we have LLMs generating responses and now we want to move on to a world where actions need to be taken. And when the problem that we are working on is a complex one with multifaceted aspects associated with it, that's where multi-agentic comes into place. So basically, we have a large complex goal, which we have to break down into specific steps and each step is basically narrowed to a specific agent.”

• •

  Chat Concierge streamlines the auto dealership experience

  “Chat Concierge is essentially an auto dealership project or application that was deployed out to our auto dealers to basically bridge that experience between dealers and their customers and make it very seamless. This is an auto buying experience that we wanted to make sure that we deliver the right solutions or cars to the right customer needs. It was a multi-agentic chat experience that was brought to the fore with the human in the loop to car buying customers to get the right match.”

• •

  Planner agents manage intent disambiguation and reasoning

  “In this specific scenario, there were a multitude of intents, so there had to be one agent that understands specifically this intent and tries to disambiguate by asking clarifying questions back to the customer. That is that narrow job. And then from there on, we had multiple tools that can get executed in the form of different actions that need to be taken based on the intent that comes in. So we have a planner agent that does this discernment.”

• •

  Platforms separate agent design from runtime governance

  “These platforms come into the fore when you are governing agents in runtime. And that's where the massive huge benefit of platforms comes into the fore. This gives the architects of that specific agentic framework the flexibility of focusing deeply on the design, whereas the platform brings in all of the governance and risk compliance that needs to be bounded to make these agents execute safely in any environment.”

• •

  Model risk frameworks are embedded in agent platforms

  “We have a very, very robust model risk office that we work very, very closely with. We have all of the risk and compliance frameworks embedded within the platform, which appear as policies, as guardrails, as security enforcement, and cyber enforcement across our different layers of the platform that get implemented across different threat boundaries of the agents.”

Episode Description

In this episode, Rashmi Shetty, senior director of enterprise generative AI platform at Capital One, joins us to explore how the company is designing, deploying, and scaling multi-agent systems in a highly regulated environment. Rashmi walks us through Chat Concierge, a multi-agent chat experience for auto dealerships that handles intent disambiguation, tool invocation, and human handoffs to deliver safer, more personalized customer journeys. We discuss Capital One’s platform-centric approach to AI agents and how it separates design from runtime governance, embedding policies, guardrails, and cyber controls across agent threat boundaries. Rashmi shares how the team approaches the developer experience for agent builders, observability, and evals for stochastic, multi-agent workflows; and strategies for model specialization, including fine-tuning and distillation. We also cover standards and abstraction, closed-loop learning from production telemetry, and key lessons for enterprises building agentic systems. The complete show notes for this episode can be found at https://twimlai.com/go/765.