Building a human-in-the-loop Slackbot agent (mays.co)

A new human-in-the-loop Slackbot agent reference app has been developed to manage financial transactions while mitigating risks associated with automated decision-making. This app screens seller names against the U.S. Treasury’s OFAC sanctions list before approving payouts, ensuring compliance by employing a tiered decision-making process that includes automated logic, machine learning models, and human reviewers. The app runs on AWS infrastructure and interacts with Stripe for payments, Slack for notifications, and the Claude API for decision support, demonstrating an innovative approach to transactional safety in fintech environments. This project is significant for the AI/ML community as it highlights practical applications of integrating human oversight in AI decision-making systems, addressing common failure scenarios such as model output errors and dual-write problems between databases and messaging systems. By implementing a transactional outbox pattern and robust error handling mechanisms, the app ensures funds are managed securely while allowing for scalability. The reference app serves as a valuable resource for developers interested in building similar systems that balance automation with necessary human control, ultimately pushing forward the conversation on safe AI practices in compliance-heavy industries.