LendingTree builds AI mortgage assistant on Amazon Bedrock

LendingTree built a multi-agent mortgage assistant on Amazon Bedrock to tackle the complexity of home financing decisions. The system uses three AI agents—a supervisor and two specialized workers (education and matching)—coordinated via LangGraph and the Model Context Protocol (MCP). All agents run on Amazon ECS with AWS Fargate, with Amazon Bedrock providing multi-model flexibility and inherited governance controls. The solution prioritizes compliance with mortgage industry regulations through Amazon Bedrock Guardrails for content filtering and PII redaction, plus a concurrent safety classifier based on a large language model.

The architecture features a React-based consumer interface served through Amazon ECS with Fargate, while the Supervisor agent acts as an orchestrator using a state machine pattern. It analyzes user intent with Amazon Nova Pro and routes requests to specialized workers or answers directly. All paths through the system are auditable and traceable, ensuring transparency. The multi-agent approach enables LendingTree to handle complex borrower questions while matching users with competitive mortgage offers.