LoanPro Agentic AI Assistant
2 steps1 pieces3 Sources
LoanPro, a US-based API-first lending platform serving over 450 lenders, has deployed a multi-agent AI assistant in collaboration with AWS and partner AllCloud to streamline loan servicing. The system addresses the operational bottleneck where human agents previously spent minutes manually navigating multiple screens to synthesize approximately 1,400 account fields from eight API endpoints during borrower calls.
The architecture utilizes Amazon Bedrock AgentCore and Anthropic’s Claude models (Sonnet, Haiku, Opus 5.5) orchestrated via the Strands Agents SDK. A composer agent routes requests to specialized sub-agents for data retrieval or action recommendations. Crucially, tool access is isolated through an MCP Gateway, ensuring that internal APIs are protected by role-based access controls and signature validation against LoanPro’s SSO service. This design allows individual agents to be tested and tuned independently without full system redeployment.
Early results reported by the vendor indicate customer call times reduced by up to 15% and account summary generation cut from 3–5 minutes to under 15 seconds. Support escalations decreased by nearly 8%. The solution emphasizes compliance in regulated credit scenarios by logging all AI actions to an audit trail attributed to named individuals. While the vendor claims a 99% accuracy target on expert-labeled benchmarks, these figures remain self-reported and have not been independently verified by third-party testing.
How it got here
2026-11-17: AWS Machine Learning Blog publishes co-written post detailing LoanPro's agentic AI architecture using Amazon Bedrock AgentCore.
aws.amazon.com ↗2026-11-18: DisconfirmAI editorial highlights the compositional routing and MCP gateway security boundaries in the new assistant.