How JP Morgan Built An AI Agent for Investment Research with LangGraph

David Odomirok and Zheng Xue from JP Morgan Chase Private Bank describe how they built “Ask David” – a sophisticated multi-agent AI system designed to automate investment research for thousands of financial products.

With billions of dollars in assets at stake, this isn’t just another chatbot – it’s an enterprise system built with human oversight for high-stakes financial decisions.

Purpose and Goals

“Ask David” was designed to automate and enhance investment research by processing vast amounts of financial data, reducing manual effort, and enabling financial professionals to focus on strategic decision-making.

The system aims to provide precise, curated, and personalized insights for financial advisors, analysts, and due diligence specialists, addressing the challenges posed by the complexity and volume of financial data.

Architecture and Technology

Multi-Agent Architecture with LangGraph: “Ask David” utilizes a multi-agent system built with LangGraph, which allows for specialized sub-agents to handle distinct tasks such as:

• Data Integration: Seamlessly combines structured data (e.g., spreadsheets, databases) with unstructured data (e.g., documents, emails, audio recordings).
• Data Processing and Analytics: Generates actionable insights using proprietary analytics and visualization tools.
• Role-Specific Personalization: Adapts outputs based on the user’s role, ensuring relevance for financial advisors or due diligence specialists.

Built on principles similar to generative pre-trained transformers, the system uses advanced algorithms to analyze market data, predict trends, and provide tailored responses to complex queries in real time. The development process involved iterative refinement with human oversight to ensure accuracy and relevance of outputs.

Development Process

J.P. Morgan recognized the time-intensive nature of investment research, which involves sifting through extensive datasets. The goal was to automate these processes to save time and improve efficiency.

The system was developed through continuous refinement, integrating feedback from financial professionals to enhance its capabilities. LangGraph facilitated the creation of a modular, scalable system where sub-agents could independently handle tasks like data retrieval, processing, and insight generation, ensuring a cohesive workflow.

“Ask David” was rolled out to J.P. Morgan’s Asset and Wealth Management division, with ongoing enhancements to handle increasingly complex queries.

Key Features and Impact

• Real-Time Insights: The system delivers instant, personalized insights by processing both structured and unstructured data, enabling faster decision-making.
• Efficiency Gains: By automating data analysis, “Ask David” reduces the time spent on research tasks by up to 95%, allowing advisors to manage larger client rosters (targeting a 50% increase over three to five years).
• Scalability and Versatility: It supports diverse asset classes, including equities, fixed income, cryptocurrencies, and ESG investments, making it adaptable to various client needs.
• Enhanced Client Experience: The AI provides tailored investment strategies, improving client interactions and fostering stronger relationships.

Challenges and Solutions

The vast and intricate nature of financial data posed a challenge. J.P. Morgan addressed this by leveraging LangGraph’s ability to integrate and process diverse data types. Human oversight was incorporated to ensure the AI’s outputs were accurate and contextually relevant, mitigating risks of errors in financial advice. The modular design of LangGraph allowed the system to scale across different user roles and applications within the bank.

Broader Implications

“Ask David” sets a new standard for financial intelligence, blending automation with human expertise to redefine investment research. It aligns with J.P. Morgan’s broader AI strategy, which includes tools like LLM Suite and Coach AI, contributing to $1.5 billion in cost savings through efficiencies in fraud prevention, personalization, and trading analytics. The system is poised to evolve, handling more complex queries and expanding applications across J.P. Morgan’s operations, potentially influencing other financial institutions to adopt similar technologies.