Dark Matter buys House Numbers’ AI platform to automate mortgage work

Dark Matter Technologies has bought House Numbers’ AI agent platform and related intellectual property, bringing its builders into the mortgage software company. The October 30, 2026 deal is intended to put agents inside lenders’ existing workflows, handling paperwork and borrower communications while people supervise. Financial terms were not disclosed.

House Numbers co-founder and CEO Jeff Levinsohn joins as chief AI officer, responsible for Dark Matter’s AI strategy and development. Fellow co-founders Arthur Granado and Benjamin Hoffman become vice president of engineering and vice president of AI adoption and enablement, respectively. The platform’s engineering and product team is also joining.

The purchase grew out of partnership discussions. CEO Vikas Rao told HousingWire he proposed an acquisition so Dark Matter could integrate the technology deeply rather than offer it as a separate service. The intended destinations include Empower, its system for managing mortgage applications through origination, the built-in borrower-facing point of sale, and other products.

Agents that work through the loan file

Dark Matter says the acquired platform has been in production for two years and has processed loans for more than 100 lenders. Rao said House Numbers has 19 agentic AI use cases—tasks software agents can carry out—in production, with more applications in development.

The platform monitors loan files for changes, then takes actions, makes recommendations or contacts borrowers and lenders by email, text and voice. Loan operations staff supervise that work. Dark Matter described several tasks the technology is designed to handle:

• Help borrowers complete applications, explain outstanding requirements and provide feedback when submitted documents are incorrect or incomplete.
• Review incoming documents to determine whether they satisfy loan conditions, and check closing packages for missing pages or signatures.
• Send reminders and flag potential problems for loan officers, helping move files through the origination process.

The hard part of agentic AI is not getting a model to do a task once. It is orchestrating complex agentic work safely inside a regulated process and executing valuable tasks consistently at scale, with people in control of the decisions that matter.

Jeff Levinsohn, incoming Dark Matter chief AI officer, in the acquisition announcement

Lenders set the limits

Rao said lenders will determine when agents can act independently, make recommendations or flag items for human review. They will also be able to train agents to follow their own policies and procedures. He described the goal as having agents perform tasks immediately, with humans reviewing the work—not eliminating people from loan fulfillment.

According to Dark Matter, the platform limits each agent to the information and capabilities its task requires. External AI assistants can access loan files through a gateway using the same permissions and controls as human users. A code-enforced compliance blocklist constrains agents, and an audit trail records actions taken by both people and software.

Dark Matter plans to run the acquired technology within its Amazon Web Services infrastructure, with controls intended to address security and regulatory requirements.

A rollout measured in weeks, savings still ahead

Dark Matter expects customers to begin receiving new capabilities within weeks of the announcement, followed by additional releases over the next several months. The House Numbers team will work across the company’s products, including AI tools already in production. That timetable describes planned integration releases, not the availability of every acquired capability today.

Rao’s financial case centers on mortgage operating costs. He estimated that producing a loan costs about $12,000, including roughly $6,000 in operational expenses. Saving even $1,000 from that operational portion would be “game changing,” he said, adding that he expects the eventual technology package to exceed that saving. Those are his estimates and expectations, not measured results from the integration.