Clearwater Analytics: Leading the AI Revolution in Finance with Multi-Agent Systems

At Clearwater Analytics, we’ve always been at the forefront of leveraging technology to drive business solutions in the financial sector. Recently, we took a monumental step in the world of AI by becoming the first in the industry to implement a Multi-Agent AI Finance System—a groundbreaking achievement that integrates Retrieval-Augmented Generation (RAG), fine-tuned models, and a multi-layered security approach that includes Galileo Protect and Azure Content Filtering.

In collaboration with Predibase and Galileo , Clearwater Analytics showcased this revolutionary system during a demo-driven webinar that outlined the end-to-end architecture and workflow of building, evaluating, and deploying multi-agent systems at scale. This article will explore the key innovations behind this achievement and what it means for the future of AI in the financial industry.

A New Era of Multi-Agent AI Systems

Multi-agent AI systems are designed to work collaboratively, where each agent performs a specific task or operates in a specialized domain. By leveraging fine-tuned models, each agent in our system can handle highly specialized tasks with increased speed and precision. This architecture enables our system to deliver fast, cost-efficient, and highly reliable results for our clients, ensuring that they can streamline their processes without compromising on quality or accuracy.

But the real power of multi-agent systems comes from their ability to interact seamlessly with one another. Each agent's output can serve as the input for another, creating a cohesive ecosystem that addresses complex problems autonomously. Whether it's analyzing vast amounts of financial data, generating reports, or offering real-time insights, these agents work in tandem to ensure tasks are completed swiftly and efficiently.

How We Built and Deployed Our Multi-Agent System

Clearwater's multi-agent system is architected around fine-tuned SLMs (small language models), which are optimized to perform specific financial tasks. This specialized approach not only improves efficiency but also minimizes latency, allowing us to reduce operational costs while providing high-speed results.

Key Highlights:

• Specialized Models: Our system utilizes small, highly-tuned models that handle specific tasks, reducing processing time and delivering results with minimal resource consumption. The following article has more details about our evaluation system for the Fine-Tuned models

• Scalable and Adaptable: Built on a scalable architecture, the system is designed to handle a growing number of tasks as our clients’ needs evolve, making it adaptable to changes in the financial sector. This article has more details about Clearwater's system

• MRKL Agents: Clearwater's MRKL agents are designed to enhance AI capabilities by integrating large language models with modular tools, enabling them to reason, retrieve real-world data, and perform complex tasks. These agents leverage a structured approach, allowing seamless interaction with multiple tools, such as calculators and APIs, to provide accurate and efficient solutions.

• RAG for Enhanced Performance: By integrating Retrieval-Augmented Generation (RAG), the system ensures that the agents always have access to the most relevant data, allowing them to produce accurate and informed outputs. This article has more details about our RAG system

Security First: Protecting Data with Galileo Protect and Azure Content Filtering

When it comes to AI in finance, security is paramount. Financial data is highly sensitive, and any potential breach can have catastrophic consequences. That’s why we incorporated Galileo Protect and Azure Content Filtering to ensure that our multi-agent system operates in a secure environment.

Galileo Protect monitors the behavior of the AI models in real-time, safeguarding against potential risks such as model hallucinations or inaccurate outputs. Meanwhile, Azure Content Filtering adds another layer of protection by ensuring that the system adheres to content and data policies, blocking any unauthorized access to sensitive information.

This dual protection method ensures that every interaction within our multi-agent system is secure, compliant, and reliable. Our clients can rest assured that their data is safe and their financial insights are accurate.

Evaluating, Monitoring, and Protecting GenAI Applications

Beyond security, we implemented tools for continuous evaluation and monitoring of our GenAI applications. By closely monitoring the system’s outputs, we ensure that any inaccuracies or inefficiencies are detected and addressed before they affect our clients. This proactive approach allows us to maintain high standards of performance while offering the agility needed to adapt to market changes.

Why This Matters for Financial Institutions:

• Minimized Latency: The system is built for speed, ensuring that clients receive timely insights without delays.
• Improved Accuracy: By using fine-tuned models and RAG, the system ensures that every result is backed by the most relevant and up-to-date data.
• Security and Trust: With built-in safeguards like Galileo Protect and Azure Content Filtering, clients can trust that their data is secure, and results are accurate.

The Future of AI in Finance

Clearwater Analytics’ multi-agent AI system is just the beginning. As we continue to innovate, our goal is to make AI more accessible, reliable, and secure for financial institutions worldwide. The success of our multi-agent architecture demonstrates that complex financial systems can be optimized through AI without compromising security or accuracy.

This milestone not only places Clearwater at the cutting edge of AI advancements but also sets the stage for future innovations. As AI models become more refined, financial institutions can expect smarter automation, reduced operational costs, and greater insights from their data.

Watch the Full Webinar

For a deeper dive into the architecture and functionality of Clearwater’s multi-agent AI system, you can watch the full demo-driven session where we explain how these systems were built and deployed at scale. Learn how we:

• Architect multi-agent systems with small, fast, specialized models.
• Minimize latency and cost through fine-tuning.
• Evaluate, monitor, and protect GenAI applications.

?? [Watch the full webinar here ]

Other articles explaining the system: