Ant Group's financial multi-agent framework is officially open source

On May 25, at the 7th Digital China Construction Summit, Ant Group officially open sourced the multi-agent framework agentUniverse. This is the industry’s first open source multi-agent technology framework in the financial field. The core of the framework provides multi-agent collaborative orchestration. Components allow developers to develop and customize multi-agent collaboration models, which can help developers speed up the research and development of large model technology in financial scenarios.

(Picture: Lu Xin, General Manager of the Intelligent Services Department of Ant Group’s Wealth Insurance Business Group, released the first open source multi-agent framework for financial scenarios)

Agents can simplify the interaction between users and large models, allowing users to only specify Goals can drive large models to complete tasks. According to the "2024 AIGC Top Ten Trends in Application Layer" report released by IDC, agents will be the mainstream form of large-model implementation business scenarios.

The technology surrounding intelligent agents is also evolving rapidly. Ant's open source agentUniverse framework is designed to help developers quickly and efficiently call any large model capabilities in financial technology and other fields to build final application products.

The core of this framework provides the multi-agent collaboration orchestration component "PEER", which is equivalent to a pattern factory, allowing developers to develop and customize multi-agent collaboration modes, and also comes with all the keys to building a single agent. components. The framework is pre-configured with various multi-agent collaboration model components that have been proven effective in real industries. Developers can easily build multi-agent applications based on this framework, and communicate and share model practices in different fields through the community.

This is also the industry’s first open source project for multi-agent in financial scenarios. agentUniverse introduces the expert knowledge framework mechanism for the first time, which reflects the integration of artificial intelligence and expert intelligence and can better adapt to more complex application scenarios. The cooperation process between multiple agents has become intelligent and can iterate independently. optimization.

In the financial industry, in-depth analysis scenarios often involve multi-level analysis, disassembly and step-by-step execution, such as interpretation of events, macro- and meso-level analysis, feasibility analysis of business plans, etc., which requires intelligent agent linkage and collaborative execution. .

"PEER" is the most distinctive multi-agent collaboration model component in the agentUniverse framework. This model uses four agents with different responsibilities: Planning, Executing, Expressing, and Reviewing. Achieve multi-step dismantling and step-by-step execution of complex problems, and conduct independent iterations based on evaluation feedback, ultimately improving the performance of reasoning and analysis tasks.

PEER model is essentially a simulation of the hierarchical cooperation and teamwork of human financial experts: the planning agent (Plan) plays a leadership role, breaking down the original complex problem into multiple sub-tasks that need to be executed separately. The remaining three agents each have their own strengths and are dependent on each other in terms of workflow. The work results are delivered in sequence, avoiding resource competition. The final work results are passed back to the "leader" for decision-making, whether the task has been completed, needs to be supplemented and optimized, or overridden and redone.

Lu Xin, general manager of the Intelligent Service Department of Ant Group’s Wealth Insurance Business Group, said that the PEER model is a sharp edge for solving inference and analysis tasks.

According to Lu Xin, the agentUniverse framework is the latest exploration result of Ant Group in applying large model technology to financial scenarios. It has been verified to be effective in real scenarios. Two typical applications include Zhi Xiaobao and Zhi Xiaozhu.

Zhi Xiaobao is a consumer-oriented application product. It is targeted at financial management and insurance professional knowledge Q&A. It also provides professional tools such as market and position analysis, intelligent underwriting and claims settlement. Currently, ZhiXiaoBao version 2.0’s ability to analyze and reason about financial events has reached the average level of real-person industry experts, and has served more than 43 million users.

Zhixiaozhu is a product series for financial professionals. It is committed to improving the professional level and production efficiency of experts in various functions of the financial service chain, and providing services in different scenarios such as investment, financial management, and insurance. Among them, the "Zhixiaozhu Analyst Edition" has fully covered important listed companies and fund products in the Chinese/U.S./Hong Kong markets, 800+ industry sectors, 200+ hot events, and more than 100 core unicorn companies, achieving financial analysis. The minute-level analysis response and multi-angle analysis of market dynamics have reached cooperation with a number of leading financial institutions, greatly improving the work efficiency of hundreds of financial analysts and financial consultants.

The above is the detailed content of Ant Group's financial multi-agent framework is officially open source. For more information, please follow other related articles on the PHP Chinese website!