Simplifying Mortgage Underwriting with Specialized AI Systems
Friday Harbor, an AI-first startup supported by the AI2 Incubator, sought to improve mortgage underwriting efficiency. By using a multi-agent AI framework, Friday Harbor aimed to automate underwriting tasks by breaking them into manageable stages, each handled by a dedicated AI agent. We partnered with Friday Harbor to design one of the AI agents and integrate it into the broader system, helping the company finish their MVP.
With a compelling business case — addressing inefficiencies in mortgage underwriting through AI — Friday Harbor secured financial backing from the Allen Institute for AI (AI2), a nonprofit founded by Microsoft co-founder Paul Allen in 2014.
The AI2 Incubator’s support enabled Friday Harbor to develop a multi-agent AI system based on segmenting the underwriting process into distinct, manageable stages. Each stage, traditionally performed by human underwriters, would be handled by a specialized AI agent, forming a modular system customized to lenders’ needs.
The project called for a combination of advanced software development, AI expertise, and deep financial domain knowledge. The critical part involved analyzing the underwriting process in detail to define each agent’s role, determine how outputs would connect, and establish a seamless handover process between agents.
.webp)
.webp)
Scope of Work: Building a Specialized AI Agent
The development required a blend of expertise to determine where traditional calculations could suffice and where AI had to step in to deliver accurate and reliable outcomes aligned with the use case.
Decomposing the Underwriting Process into Stages
After the traditional workflow used by human underwriters was broken down into distinct, actionable tasks, we identified the unique role each AI agent would play in returning accurate and dependable results.
Developing a Specialized AI Agent
We designed and implemented one of the agents in the multi-agent system, equipping it with the necessary tools and logic to perform its designated tasks within the network.
Integrating the Agent into the Multi-Agent System
Effective communication between agents required careful integration. This step allowed outputs from one agent to serve as inputs for others, maintaining data consistency throughout the workflow.
Scope of Work: Building a Specialized AI Agent
The development required a blend of expertise to determine where traditional calculations could suffice and where AI had to step in to deliver accurate and reliable outcomes aligned with the use case.
Decomposing the Underwriting Process into Stages
After the traditional workflow used by human underwriters was broken down into distinct, actionable tasks, we identified the unique role each AI agent would play in returning accurate and dependable results.
Developing a Specialized AI Agent
We designed and implemented one of the agents in the multi-agent system, equipping it with the necessary tools and logic to perform its designated tasks within the network.
Integrating the Agent into the Multi-Agent System
Effective communication between agents required careful integration. This step allowed outputs from one agent to serve as inputs for others, maintaining data consistency throughout the workflow.
.webp)
Solution
The multi-agent AI system was designed with modularity in mind, enabling each agent to perform specialized underwriting tasks independently while contributing to the shared goal of generating accurate and fully underwritten loan files.
Features such as real-time document analysis and integration with existing systems help in more efficient loan file assembly, clearing conditions faster and with greater accuracy.
Nomtek’s role was building and integrating one of the agents, aligning its functionality with the broader system to create a cohesive network. This modular approach also allowed for adaptability, making it easier to update or fine-tune individual agents as the system matured.
The project advanced Friday Harbor’s vision for a scalable, AI-driven mortgage underwriting solution. By breaking down complex processes into smaller, manageable components, the multi-agent system demonstrated potential for reducing manual oversight and improving efficiency in financial underwriting. The work laid a strong foundation for further development, helping Friday Harbor demo the MVP and showcase its underwriting capabilities.
.webp)
.png)
%201.png)
