While many startups promise world-altering AI agents, Kensho, a division of S&P Global, is taking a pragmatic approach. Instead of chasing speculative AI applications, they are addressing a critical challenge: ensuring AI operating on financial data uses verified information. Kensho's objective is to anchor AI outputs to trusted sources, regardless of how that information is ultimately used. This focus on reliability is paramount in finance, where errors can have severe consequences.
The data managed by S&P Global is not simple text suitable for a standard LLM. It comprises highly structured and nuanced datasets that require a specialized approach, rendering conventional search engines insufficient. To streamline access to this data and integrate it with advanced generative AI applications, Kensho has developed a framework called Grounding. This framework will serve as a unified entry point to S&P Global's data, ensuring that all AI-generated insights are based on verified datasets. This is key to unifying data from different business units and integrating it with AI agents, LLMs, and GenAI applications while maintaining a high degree of trust and compliance.
Financial professionals are fatigued by data fragmentation, spending hours on search and verification. Grounding aims to solve this by providing a single access point for natural language queries to S&P Global's verified financial datasets. You can forget about complex database schemas and specialized query languages. The system will deliver accurate, real-time information, citing Kensho's trusted sources. This allows users to concentrate on analysis rather than the endless pursuit and validation of data.
To implement this, Kensho is leveraging LangGraph to build a multi-agent architecture. Grounding acts as the central access point for AI agents, which in turn retrieve data from various S&P Global sources. Rather than embedding natural language processing logic within each agent, developers have created a router. This router intelligently directs queries to specialized data retrieval agents (DRAs). These agents, assigned to different domains such as equities, fixed income, and macroeconomics, facilitate a division of labor and enhance the efficiency of working with highly specialized data.
Creating multi-agent frameworks like Grounding is becoming critically important for any industry where data accuracy and reliability are non-negotiable. For the financial sector, medicine, law, and engineering, this signifies a shift from hyped AI promises to the actual automation of complex processes based on verified information. Ultimately, this enhances efficiency and mitigates risks.
Why this matters: Businesses in data-sensitive sectors must prioritize AI solutions that ground outputs in verified information to build trust and ensure operational integrity. Consider implementing frameworks that centralize data access and verification for your AI initiatives.