Clause extraction and classification: ISDA

The Arizona State University Artificial Intelligence Cloud Innovation Center (AI CIC), powered by Amazon Web Services (AWS), collaborated with the International Swaps and Derivatives Association (ISDA) to address the challenge of automating the extraction and classification of clauses from the ISDA Credit Support Annex (CSA), a key document used in the derivatives market to manage credit risk between parties involved in derivative transactions. This collaboration aimed to streamline the complex process of identifying and categorizing clause variants, ultimately presenting the information in a JSON format that aligns with the Common Domain Model (CDM), a framework developed to standardize and streamline the representation of financial transactions and their associated data.

CSAs are a key component in the derivatives market, essential for managing counterparty credit risk, while the CDM is crucial for standardizing and improving the efficiency of financial transactions. However, the process of extracting and classifying clauses from legacy CSAs is time-consuming and prone to errors. The challenge was to develop a Proof of Concept (POC) that could automatically extract clauses, identify variants, and present this information in a structured JSON format.

The AI CIC team employed a multi-phase approach to develop a scalable solution using AWS cloud services. The project involved creating an agent architecture that leveraged prompt engineering techniques to accurately extract and classify clauses from the ISDA Master Agreement. The solution utilized Amazon Bedrock, AWS Lambda, and AWS S3 to manage document uploads, trigger clause extraction processes, and store the resulting CDM JSON structures.
Key milestones in the project included the development of specialized prompts for each clause, testing various large language models (LLMs) to determine the best fit, and integrating a bedrock agent that could handle the extraction and classification tasks. The Claude 3 OPUS LLM was selected for its superior performance in reasoning, classification, and handling complex prompt instructions.
 

The successful implementation of this project demonstrates a significant advancement in automating legal document information extraction. By leveraging AI and cloud technologies, the project provides a scalable and efficient solution for the derivatives industry, reducing manual labor and minimizing the risk of errors. This automation not only enhances the accuracy of clause extraction but also enables faster decision-making processes, ultimately contributing to more robust and stable financial markets.

"I was introduced to the Generative AI work at the Cloud Innovation Center by our AWS contact and had the privilege of collaborating with them to build a proof of concept (PoC) for an AI process that converts legacy legal documents into our common domain model format. I want to extend my gratitude to Arun, Sai, and the entire team for their excellent work in developing the PoC. This achievement allows us to further develop and promote the Common Domain Model and facilitates the integration of legacy documents using the CDM. The team's work was fast and professional, making the experience outstanding. I highly recommend this program to other organizations in the public sector."

David Lee, Senior Director, Digital Transformation,International Swaps and Derivatives Association, Inc. (ISDA)

The techniques and architecture developed in this project have the potential to be applied across various industries where legal document processing is a critical task. The solution can be adapted to other complex agreements, such as contracts in finance, real estate, and corporate law, providing a powerful tool for organizations to automate the extraction and analysis of critical information.