Various German Banks
BankAI: Reducing Contract Risk with Self-Hosted GenAI
GenAI solution for document generation and validation, such as investment and loan agreements, significantly reducing errors and operational risks.
Objective
Minimize contract errors to reduce financial and regulatory risks through AI-driven document validation.
Challenge
Creating a reliable GenAI model capable of high accuracy and compliance within regulated banking environments, overcoming complexities associated with developing and implementing a tailored financial-sector model before similar technologies like ChatGPT become widely available.
Solution
Self-hosted GenAI solution using TensorFlow, an open-source machine learning library. Developed native client plugins for Microsoft Office and LibreOffice. Implemented a module for integrating TensorFlow with the bank's existing Document Management System (DMS).
Results
Reduced contract errors by over 70%. Decreased the time required for creating large contracts from 16 hours to just 4 hours. The solution was acquired by Deutsche Bank for $2.5 million in 2023.
Tech stack
TensorFlow, Keras, PyTorch, Apache Airflow, Python, Azure, JavaScript, React Native, React.JS, C++, Microsoft Office Add-ins Development Kit, LibreOffice Extensions, MongoDB, GitLab CI, OpenSearch