• We used Amazon Bedrock as the primary model provider for customer and underwriter explanations, with Bedrock Knowledge Bases supplying policy context. The architecture also includes AgentCore and Strands Agents SDK foundations for orchestration, deployment, identity, and observability.

  • We used TinyFish Search, Fetch, and Agent APIs for the Policy Sentinel workflow. It discovers approved public policy sources, retrieves relevant guidance, navigates complex pages when needed, and converts findings into structured evidence. That evidence supports draft policy proposals, which are then simulated by the existing credit engine and reviewed by humans before approval for testing.

Built With

Share this project:

Updates