LendingTree builds multi-agent mortgage assistant on Amazon Bedrock
Amazon Web Services
LendingTree has deployed a multi-agent mortgage assistant on Amazon Bedrock, using LangGraph and MCP to coordinate three agents. The system handles educational and matching tasks, with production analytics showing over 97% containment and growing transactional use.
LendingTree built a multi-agent mortgage assistant on Amazon Bedrock to help consumers navigate home buying. The solution uses three independent AI agents—a supervisor and two specialized workers (education and matching)—coordinated via LangGraph, the Model Context Protocol (MCP), and foundation models from Amazon Bedrock. All agents run containerized services on Amazon ECS with AWS Fargate. The supervisor analyzes user intent with Amazon Nova Pro and routes tasks using MCP, while the education worker uses Amazon Bedrock Knowledge Bases for retrieval augmented generation, and the matching worker calls internal APIs for personalized offers. Production analytics from late 2025 through Q1 2026 show the assistant handled about 1,960 conversations and 12,100 messages, with engaged users averaging 10+ messages over 9 minutes. Over 97% of conversations were contained without human escalation, and recent conversations show over 50% transactional intent such as rate comparisons and lender matching.
- Abbreviations
- MCP = Model Context Protocol — Протокол контекста модели
- ECS = Amazon Elastic Container Service — Эластичный контейнерный сервис Amazon
- PII = Personally Identifiable Information — Личная информация, позволяющая идентифицировать личность
- LLM = Large Language Model — Большая языковая модель
- RAG = Retrieval Augmented Generation — Генерация с дополнением поиском
- CI/CD = Continuous Integration and Continuous Delivery — Непрерывная интеграция и непрерывная поставка
Source: AWS ML blog —
original