This founder built the AI that runs mortgages for America’s two biggest homebuilders

^{A week in AI is like a year in other industries. I hope these issues become your weekly source of AI information, inspiration, and ideas. If we haven’t met before, I’m Amanda Smith. I write about AI and the fascinating folks who are building in this brave new world.}

After a massive exit, this week’s founder didn’t spend the rest of her days chilling at the beach. She went on to fix the most expensive and stressful purchase of your life. As an immigrant, she had personally struggled with this problem, which inspired her to get back to building. 

Let’s dive in. 

Subscribe | Past essays | Let’s collaborate | Insane Media

🤝 This edition is kindly brought to you by Infisical

If you’re building agents that call internal APIs, databases, or MCP servers, this live technical walkthrough shows how to keep credentials out of agent context, restrict access to approved actions, issue short-lived session tokens, and log every tool call. You’ll leave with patterns you can apply to your own agent infrastructure, plus a demo of Infisical Agent Vault.

Save your spot

Fax machines, printed bank statements and loan files with 500+ pages: the mortgage behind the biggest purchase of your life still runs on 1990s technology. 

Diane Yu has been through that process twice as a borrower. The first time, she was a new immigrant still learning English. The second came years later, when she applied for a mortgage again after a successful exit. The process was just as painful. 

As CTO of digital lender Better.com, she saw that lenders wanted to do better but had no good tool. She could have advised or invested. "I'm an engineer. When I see a problem, I have to solve it, or I can't sleep."

So she started over as a founder again and started building Tidalwave, an AI-powered end-to-end self-serve mortgage platform. 

"People inside mortgage see the manual work as part of the job. I saw a system design problem."

A mortgage takes about 43 days to close. Lenders spend about $11,000 to originate a loan and clear $973 in profit. Over the past decade, 11.5% of loan file content has been wrong or missing. "To an engineer, that means the technology isn't doing the work. People are."

Most mortgage software assumes a simple borrower. Add a co-borrower, a gift from parents or a second job, and the file goes back to a human.

Tidalwave's agents do a loan processor's work: reading documents, checking income and assets, running loans through Fannie Mae and Freddie Mac's systems, and chasing borrowers for what's missing. A forensic bank statement review that takes an underwriter two to four hours takes the agent five or six minutes, with every flag linked to the exact line.

Tidalwave is model-agnostic and doesn't train its own foundation model. It built the Mortgage Contextualizer, a patent-pending knowledge graph of every document, transaction and finding in a loan. It feeds the language model precise context, strips out personal information and is tuned to each lender's rules. 

Yu said they built a lot of it next to customers. First Colony Mortgage spent a year co-building it on 700-page loan packages. This was key in a highly regulated industry. 

"When a chatbot gets something wrong, usually nobody gets hurt. In mortgage, a wrong answer is a bad loan or a compliance violation, on a family's biggest purchase."

On the first public benchmark of AI on real mortgage tasks, run with Columbia University's DAPLab, Tidalwave's agent scored 95% on compliance checks against 42% for a general-purpose model. It scored lower on account verification because it would rather leave a transaction out than make one up. "That's the failure mode I want in a regulated product." A human still makes the final call.

Independent lenders came first, then homebuilders, for whom a slow mortgage means a finished home sits empty. "If it works at that scale, it works anywhere."

Today, Tidalwave runs the lending arms of the two largest homebuilders in the country, DHI Mortgage and Lennar Mortgage, plus retail lenders, wholesale lenders, banks, and credit unions.

They achieved this by asking for a pilot, then earning the rest. 

Tidalwave plugs into existing systems, goes live within 90 days and passes big compliance reviews ("You can't vibe-code your way past them"). Nine staff came from ICE Mortgage Technology, whose systems most of these lenders use. When we sit across from a builder’s lender, we speak their language.” 

The strategy is to land an anchor in every type of lender so each segment's playbook speeds up the next deal. Tidalwave has signed 31 customers, 22 of them live, and every customer up for renewal last year renewed. It's on track to touch more than 200,000 loans a year, which Yu says is 4% of the U.S. market.

Yu’s co-founders both came from her past endeavours and she said you cannot buy 20 years of trust. 

Yu tries to talk new hires out of joining. " In the first conversation we talk down, not up. You'll be paid less cash. You'll work harder. You'll take less vacation. If they keep coming back, that's my person.” For investors, she looks for people on a longer clock than one fund cycle. 

Her advice for founders deciding what's next:

If one of these stories stuck with you, I’d love to hear which one.

Speak soon,
Amanda

P.S. If you want to get a founder feature about your own story, reply to this email. If you’d like to reach our newsletter audience (founders, creators, and marketers), click the button below.

If you’re new here, I’m over the moon you’ve joined us! To help me craft content that’s actually useful (and not just noise in your inbox), I’d love it if you took 1 minute to answer this quick survey below. Your insights help shape everything I write.

Dive into our other newsletters - where psychology meets the creator economy, e-commerce marketing, and Human resources.