How Kensho built a multi-agent framework with LangGraph to solve trusted financial data retrieval

Kensho, S&P Global’s AI innovation engine, developed a multi-agent framework called Grounding to solve fragmented financial data retrieval. Built using LangGraph, the system acts as a unified access layer, routing natural language queries to specialized Data Retrieval Agents (DRAs). This architecture ensures that insights are derived directly from verified, structured datasets, providing citation-backed responses for high-trust applications. The framework relies on a custom data protocol for consistent agent communication and emphasizes observability and multi-stage evaluation to maintain reliability and compliance. This approach accelerates the deployment of specialized financial AI products by providing a robust, shared data foundation.