4kd.ai

Craig (00:00)
Hey, here we go. Another episode of Forked AI. And I'm excited today because we're going to do some show and tell, which you do not want to miss with our guest, Steve White. got our great co-host on the show today. Again, Mr. Moses. Good to see you, sir. Again, in the blazer, feeling very underdressed, as always, against you. So thanks.

David Moses (00:23)
It's good to see you.

Craig (00:25)
Jack unfortunately cannot be here with us today. He was stricken with allergies and lost his beautiful voice last night. it's Steve. Steve does not have nearly the illustrious amount of hair that Jack does, but we're going to have him on the show anyway. Steve, welcome to the show. I'll just give a quick intro. Steve has been with Dominion for a long time. He's an underwriter. So when a deal comes in from a borrower,

It's probably going through Steve's hands or Steve's team. And over the last several months, Steve has developed what we think is one of the coolest pieces of software, literally from scratch, not a coder per se, but as has a tech background. And I can't wait to show you guys the Dominion underwriting console. Dave, I think this is going to blow your mind and I can't wait to talk about it. So, Steve, welcome to the show, sir. It's great to have you.

Steve (01:18)
Thank you very much. I don't know if I can live up to that intro. That was quite an intro. Thank you. ⁓

Craig (01:24)
So Steve,

we'll get into the software in a minute, but I want to talk about, just quickly, what was your job description like at Dominion? And then also what your background is. I think you told me, not really a programmer, but you've had a tech background. So just give us a little bit of flavor, and then we'll jump into the software.

Steve (01:46)
Okay, so yeah, I am a coder, right? I mean, previously, I consider that like one of my past lives. So I used to be a software developer at Microsoft back in the 90s when that was cool.

Craig (02:00)
thought you were more

of a systems guy, not a software guy.

Steve (02:03)
No,

I was writing code. actually wrote my claim to fame is I wrote a tiny little piece of Microsoft Excel and it shipped for a long time. I don't think it's actually in the product anymore, but I did write a little wizard for it.

Craig (02:11)
⁓

Steve (02:18)
it was quite a time back in the nineties, you know, to be in software. And we had the kind of the dot com boom. If you remember that it was like the wild west, right? VC money was just pouring into the space. There was tons of wall street money everywhere. They were hiring. Yeah. Yeah. Yeah. So just outside Seattle.

Craig (02:34)
Were you in Washington or were you living somewhere else? You did? Okay.

Steve (02:40)
Yeah, was Sammamish actually, which was part of Redmond at the time, which is where Microsoft is headquartered. So it was quite a blast. Really, it was a lot of fun back then. And then what happened was the dot com crash happened and all the money dried up overnight, like in the early 2000 aughts. And I found myself laid off from a software job that I had at the time.

And I was kind of looking for a job in software and I was competing against people, you know, essentially who were prior to that, they were like janitors and waiters and waitresses and things like that. And they took one college class on how to be a software developer because of the, you know, the.com boom. They were paying so much money. was so much demand for software developers.

Craig (03:17)
Duh.

Steve (03:30)
that everybody was a software developer, right? It was like the whole world became software developers. so being out of a job around then, was like, now competing against these people for jobs who were, they were just like, well, we'll just go back to rates, really low rates that, because it's still better than going back to being a hairdresser or whatever they were before.

So, so I retired from that career and I got into real estate and, started doing real estate investing. And of course that was before the big run up to the, the, yeah, the 2008 crash. So that worked out really well for me because yeah, well, it actually worked out really well because, the, I sold my last investment property in 2007.

Craig (04:02)
2008 crash.

Your sense of timing is uncanny, Sarah.

David Moses (04:17)
Wow.

Steve (04:18)
Before the crash, had a sort of a portfolio built up, sold everything in 2007 because the prices were just so high. It was not because I was a genius. I didn't predict the future. just thought, you know what? This is fantastic money. I'm going to sell it all. I sold it all. like a month later, the market crashed. But the reason I'm telling you this story and why it's interesting is because I can see echoes of that. That whole story, I can see echoes of that today.

Craig (04:26)
Right, of course.

Steve (04:45)
So you have all of sudden.

Craig (04:46)
Steve, Steve

2008 was it all because of the exotic mortgages Steve. This is a totally different time. I mean, how can it be anything like 2008 Steve? That's.

Steve (04:53)
I'm not talking

about the real estate. I'm talking about the software because now everybody's a software developer. Right.

Craig (05:01)
⁓

David Moses (05:03)
That's true. real

Craig (05:03)
I was just being

cynical.

David Moses (05:05)
estate's not going to crash for another few years.

Craig (05:09)
Exactly. Good man. Keep going, Steve.

Steve (05:12)
So that eventually, I'll skip the next 15 years, but eventually that led me to Dominion. I've been working at Dominion in the underwriting department for roughly four years, almost four years. And when I started, everything was antiquated, like the whole underwriting system. just used literally the...

Lesson one when I first started the first day was, okay, here's how you open Microsoft Notepad on your computer and you just literally start typing your underwriting report. We'd be analyzing properties on calculators and spreadsheets and by hand scratch pads.

Craig (05:50)
And Steve,

it's not an insignificant amount of deals. Like how many deals in a month on average do you think would pass across your desk with the antiquated system?

Steve (05:59)
So in the old way. ⁓

Craig (05:59)
Yeah.

Steve (06:02)
I, but across the whole team, was hundreds. I'm not, couldn't tell you on just my own. So dozens, mean, you know, you do several per day times 30, so a hundred.

Craig (06:05)
Yeah, so.

So we're talking about

single family, multi-family, ground up construction, all of this stuff being sort of underwritten using Notepad and like a bunch of clicks into other websites. Keep going.

Steve (06:24)
Yeah, exactly. And when you say a bunch of clicks and other websites that it was dozens and dozens, you would have to open up, you know, five different websites to you'd be looking for like tax records and public.

public records about the property to do your research, your diligence. Then you would check with third party vendors. So you want to look up MLS history and ownership history, things like that. Has the property sold recently? Is it available for sale now?

Can you get photos from another vendor? And so there were many vendors of data and you'd have to open up all of them separately and go out and you're like copying and pasting into your notepad and

Craig (07:08)
You

described it as doing a massive book report on every house, like, you know, where you had to go get the microfiche from the Library of Congress, like, you know, yeah, doing that, like, you know, seven or eight times a day, like, not exactly.

Steve (07:14)
Yes.

Yeah, exactly. ⁓

Yeah, we were doing several

per day, right? Three, four, five, six, know, depending on how hard they were. Many of those per day, like a college level book report every day at Dominion. And, but we ended up with the best values, you know, I mean, that was the goal was to end up with, we're valuing the property to make intelligent investment decisions. And, and to do that correctly took time and effort and a lot of work.

Craig (07:51)
And I want to just stop you there because there's also not only the massive book report that you have to prepare. Let's talk about the flow. So customer borrower client sends us in a loan request and maybe a purchase order for the I'm sorry, the purchase agreement for the house. The loan officer then sends that loan request to underwriting.

Underwriting then assigns it to someone on the underwriting staff. Underwriting staff then produces the massive book report and sends all of that information up to Jack so that Jack can make a decent decision on whether or not we're going to fund the loan. Or maybe Jack makes some comments and says, hey, go back and find me some more information for your book report. Then let's suppose that it's approved.

That then goes to processing and the loan officer to send the quote to the borrower. And it's a lot of handoffs along the way on this workflow. And I just wanted to give that color because it's not just Steve working in his basement on an app for underwriting. is what you're you're about to see is something that will be so comprehensive and it'll be able to be communicated across all departments sort of seamlessly.

which I think is the most promising part of it all. And frankly, not even the real meat of what you've built. you know, like it's, it's a lot to it, Dave, you're going to be blown away. So Steve, any more information you want to go through and then we'll jump into it and show the audience this massive thing that you built from scratch.

Steve (09:27)
I think we're good. I think you have the idea. all I can say is this is more recent, right? So we started with Notepad. We graduated to a more Excel-based solution that I created a spreadsheet back before AI was a thing. And that helped. That certainly made our jobs easier, but not easy by any means. And then

then the AI revolution happens.

Craig (09:54)
Steve,

could you have built what we are about to see? Like, first of all, when did you start building it? How long has it been since you cut? Yeah.

Steve (10:00)
a

little over a month ago, really in earnest and ⁓ conceptually, yeah, conceptually a few months, but yeah, really working hard on it a month, maybe a month and a half, something like that.

Craig (10:04)
Really? I swear to god I thought it's been like six months.

Okay. And then,

so the whole idea here is, you know, this agentic, you know, LLM is going to do work, do hopefully all the repetitive tasks that compromise an underwrite, I'm sorry. Yeah, that compromise an underwriter's day.

And so Steve, like, you know, let's fire up the old screen share here and show folks what you built. And this is the Dominion Underwriting Console by Steve White.

Steve (10:41)
That's quite an introduction. right, just a second. Here we go.

All right, this is it. So no, not at all. It's a lot better than a spreadsheet. These are test deals, just so everybody knows. I'm not doxing anybody. These are all properties that I just picked out randomly on a map. But the gist of it is the underwriting team, everybody on the underwriting team will, this is their home page. So they will go in and ⁓ create their deals and...

Craig (10:48)
That's not an Excel spreadsheet.

David Moses (10:50)
Yeah.

Steve (11:14)
do the underwriting here. We have different types of underwriting on our team. the act of underwriting deals is where most of the action is. That's what you would think of as underwriting. We have another process called re-underwriting. There's a queue for that. That means we're just updating the records on properties that are already in our system, loans that are already in our system. And we have a couple other types, but those are the major ones up top here.

And you can see I have quite a few test deals in here. there's ⁓ some menu items. We have some administration stuff. But really, this is the meat and potatoes of it. So if I was underwriting a deal, first thing I'm going to do is hit the plus sign here, and it'll let me import it. So let me just go grab a deal. I'll find a test deal that.

That's a good one.

Craig (12:09)
So

Steve, normally when we get a loan request, it's going to be on our form. It's a PDF. Are you saying that this then extracts the PDF and sort of uploads it into your system?

Steve (12:22)
This won't do that automatically. It doesn't extract the PDF. What you're talking about is a loan request form. A lot of times, loan officers will go work out a deal with a borrower. The borrower submits a form that has all the basic info on it. It's the loan request form. That's a PDF right now.

Craig (12:29)
Thank you.

Steve (12:41)
⁓ We don't do that automatically. I'm not extracting any details from that automatically, but what will happen is the underwriter will receive that PDF and then type it into our system. We use Salesforce as a backend. And so they'll type it into Salesforce. And then when I do the import into here, it's coming from Salesforce into the tool here. So ⁓ here's what this would look like. I just imported a new deal.

Craig (12:53)
Sure.

Steve (13:10)
And up top, can see it says requirements 100%. So we have all the deal level requirements we need. We know who the loan officer is, and we have a close date for the deal. And if you expand that, you'll see a little bit more like who is your title company, what's their title email address, that kind of thing. And you can edit those details as well. ⁓ And that's about it. Up top, we just kind of have like a

sort of an intro. This is the address, this is the name of the deal. And then under that you'll have one property report per property in the deal. So if it's a multi-property deal, like if you have a six property portfolio, you'll have six of these that show up. This is a simple project with just one property, and so it's already added one property report for me, which is the address of the property.

Craig (13:52)
Mm-hmm.

Steve (14:04)
After that, you're just going to kind of go through these buttons in order to underwrite your deal. And one of the overarching sort of themes here is that I'm trying really hard to do as much work in the background for the underwriter as possible.

So before you even get here, it's kind of like in the background, seamlessly going out and downloading information from the internet for you, communicating with APIs, calling our third party vendors and downloading things for them.

David Moses (14:35)
So when you

say it's in the background, maybe I'll wait till you're done with the demonstration, actually. Go ahead. I'll ask again.

Craig (14:36)
Count us in.

Steve (14:40)
Okay.

Yeah, that's fine. So let's start out with the details. So you're just usually the underwriter is going to do these left to right. Okay, we just do them in order. You can do them in any order you want, but left to right makes the most sense. So first up, we have PDR details and

What you're going to do here is up top is I'm predicting based on things like how much budget the borrower has dedicated to rehab on a product. This is for fix and flip, of course. So the borrower is saying, I'm going to fix this up and you'll put in a description. So let me just type in complete cosmetic rehab.

with ⁓ bathroom, let's say half bath conversion to full bath. So the borrower is going to be adding a bathroom in this case. If it's a refi, it will put in a description of any rehab they've already completed. So let's say they've finished demolition.

And then I'm going to pick an expected condition and expected qualities. These match to what an appraiser would use, like the C2 rating and the Q4. We'll say it's going to be C2, which is a full renovation. And Q4, quality four, is your sort of average type of quality. And then below that, it's showing me ⁓ the details about the property from different sources.

Craig (15:54)
Yeah.

David Moses (15:54)
Ahem.

Steve (16:13)
So these are all some of our vendors for property data. The most important one is the PDR. That's the Property Detail Report. We get that from DataTree. It's mostly public records. It gives you kind of like, here's what's on the tax records. Here's the zoning. Here's the flood zone. ⁓ It'll show you previous sales. We'll see that in a minute. But you get the idea here is each column is a vendor.

And I automatically load the first two, is PDRs and House Canary. House Canary is another one of our vendors. They mostly give us MLS data. So ⁓ between those, I'm asking the underwriter, who do you believe? Which details make the most sense to you? And we're going to research using every available means at our disposal. We'll research.

Craig (16:56)
Yeah.

Steve (17:06)
you know, property detail reports, House Canary, we'll go look at Zillow and Redfin and the various public MLS type websites. We'll pull tax records from the county. We look at Google Street View. We look at Google satellite maps. know, sometimes you have a subject where everything looks great, but then you look at the satellite maps and you realize, you know, wow, it's selling for so cheap because it's next door to a junkyard and a pig farm, you know, or

Craig (17:34)
It's the

nicest house in a war zone.

Steve (17:36)
Yeah, exactly. exactly. Even street view can lie to you. Sometimes you see this beautiful street and then on the on backside of the property, there's, you know, like a strip of discos or something that's that's really loud every night. And so causing.

Craig (17:51)
Dave, you

love your discos. I know you do.

David Moses (17:52)
Are you, well,

mean, clearly by the blazer. Are you pulling the actual street view from Google or are you pulling the street view from one of these sites that's already giving you other data?

Steve (18:04)
Yeah, yeah.

Yeah, I'm not pulling the street view directly on this page, but we do pull it other places. I pull it through the API. Google has a street view API. And so with just a couple of clicks, the underwriter can get to street views and satellite views all of that.

David Moses (18:13)
Yep.

Yeah, it's

actually kind of cool.

David Moses (18:22)
Google API, the Maps API is pretty cool. For free, you can grab the metadata, which is just a fancy way of saying data about the image. That'll tell you when it's from. And then for, I think the pricing is still 7 tenths of a cent, you can pull the image. I think it's great for, you know,

What we're gonna try to use it for in the future is if we've got somebody sends us, go buy this property. We will use that to essentially scout the neighborhood for us. Right, you'll go out not just gonna get a street view of the house, but you get a street view of all of the houses within a few blocks of there. And you get a really good sense for what the area is. Because what it looks like is either where it is or where it's going.

And if the images are very old, that's why you'll look at the metadata and say, these images are from 2019. Maybe I'm not going to rely on them very much.

Steve (19:17)
That's really good tip, David. I'm going to add that to my app. I'm going to check the... Yeah, exactly. That's really good. I hadn't thought about that, but that makes perfect sense. We should check the date, right? We should check the metadata and incorporate that. Yeah, smart. So what's going to happen here on this page? The underwriter will go through and analyze which source of truth do we believe.

Craig (19:20)
we brought you on the show, Steve.

Steve (19:41)
track it down, any with using every possible means, sometimes they're going to type in their own values. And they can do that in the right column. So that's the analyst. And that's the final arbiter. You're deciding which way you think it's going to go. Sometimes you just call the borrower and ask. So in this case, I put in the rehab description of the borrower's converting the half-bath to a full-bath.

And you can see the PDR is reporting 1.1 baths. So it's got one and a half. House Canary or the MLS is showing two baths. And the analyst then could either click House Canary and say, believe that one, or they could type their own value override. Yep. And for basement types, there's no really, as far as I know, there's no really good solid vendor that provides, you know, a hundred percent.

David Moses (20:21)
That's really cool.

Steve (20:35)
⁓ reliable data on basements. that's a drop down and the underwriter will figure it out and select it from a list. After that there's other details that they can go through from the PDR. So it'll give you the last transfer date. It'll show you the last ⁓ market sale date.

And those are important because sometimes ⁓ those can be indicators of problems with the property or the title. Like if there's a bunch of recent sales, it might be something you want to look into. It might be an investor passing it to another investor, passing it to another investor. And we research those a little more carefully. But if it's a little bit of an older sale, the last sale was a couple of years ago, chances are pretty good that the new transaction that we're analyzing is

arm's length, probably not a setup kind of thing. It'll tell you about the garage and know flood zone and yeah and the underwriter has these three buttons green yellow red where they can flag anything as an issue just to

research later. let's say you were in a really tight flood zone. You could click the red there, and it would flag it as an issue. And then down here, we have a of like an auto-generated summary of all the various choices we made in the table above. It'll tell us, reconciled these different things, and these are the values we're using. Or, hey, we had this issue with the flood zone. And then the underwriter can.

add that to our commentary. So we do a thing here called a PDR review that's manual. We'll just type in sort of any issues that we discovered in the PDR review, and then we can save it. Any questions on this page?

Craig (22:20)
Very cool.

David Moses (22:21)
Yeah, that's very cool. Do you have a similar, is it completely different for DSCR versus RTL or have you not built out DSCR yet or is it, because I imagine you're really relying on an appraiser more for.

Steve (22:23)
So.

Yeah, that's an amazing question because today I'm working on the DSCR version of this. And yeah, it's very, very similar. It's going to be almost identical. Yes. Well, at some point you have to pick the source of truth for what am I looking at? And that's done in this style of table. So previously, again, this was

multiple websites you had to open up and you're copying and pasting out of each one. Now it's just effortless. Everything just shows up in a table and you can click on it, which is, you

Craig (23:08)
Steve, did you build

it with the intention that it would be scalable to DSCR or is it just so happens that you just build a great mousetrap that you can now plug into DSCR?

Steve (23:20)
Yeah, I was thinking about it the whole time. I was thinking this should work for DSCR. It might require different modifications, but that's one of the nice things about AI is you can just say, when you're doing AI software development, you can just say, oh, I'm building the DSCR version now. Please go out and copy the RTL version. This is the RTL version that we're looking at here. Residential transition.

Yeah, go out and copy that and make these four changes or whatever. It's just instantaneous. So it all works out. The next thing is requirements. So there are cases where we don't have all the requirements from the bar. Here you can see it's 84%, meaning we have 84 % of the requirements. And depending on the type of deal, we have different requirements. And this is dynamic.

So I'm looking at it and saying, okay, this is a single family home. It's a fix and flip, and it's a refinance, not a purchase. So based on those three things, what are the requirements that I need from the borrower to complete this loan quote, to complete my analysis? And I build this checklist dynamically, and then the underwriter is gonna check off each item as we receive it.

And if there's anything missing, it will generate an email for you automatically. So you can email the borrower and say, hey, I'm missing the, OK, in this case, PDR review doesn't make sense. But you can email the borrower and it'll automatically say, hey, I'm missing your assigned HUD 1 settlement statement.

David Moses (25:01)
Just not to dive too far into the weeds here, but does this send the email to the loan officer to then click to send it through, look at it and then send it through to the borrower, or does it go straight to the borrower?

Steve (25:15)
In this case, I'm using Dominion's process. Dominion, the underwriter, will contact the borrower directly. So it doesn't go through the loan officer. But that would be an easy mod to make. I could even send the email directly to the borrower on the loan officer's behalf with a reply to tag so that it goes to the loan officer.

Craig (25:36)
Do have like a template

already written for the emails and does it send it automatically or does it write it and then put it in a draft to be sent?

Steve (25:45)
Right now, it just generates text and you copy it and paste it over to your email client, whatever you use.

Craig (25:52)
Dave, you can imagine

where having Read-Write Send access for AI for the company is a little bit of a bugaboo that we're working through security-wise. I think we're very close to getting closure on that. And I think it will revolutionize the way we work without Lookaround here.

David Moses (26:11)
Yes, see

what platform are you building this all on?

Steve (26:15)
I'm using Claude code in the CLI, in the terminal.

David Moses (26:17)
Okay. ⁓ And then what,

where does it, where does the, you know, once Cloud Code writes the code, where does that code live?

Steve (26:26)
My code is automatically uploaded to Bitbucket, which is a GitHub type sort clone. And that's where that hosting system has automatic build scripts. I send, I'll say to Cloud Code, OK, we're done. Please send this to the GitHub repository. The bits get uploaded automatically to Bitbucket.

And then Bitbucket detects the incoming code automatically, runs a build script, and deploys it to production. So once I tell Claude, hey, send this to the repository, within about seven minutes, it's live on the website. Whatever changes I made are live on the website.

David Moses (27:13)
Gotcha, then so it's actually the workflow, the actual process, know, the, you know, the two in the from it's all happening in, ⁓ it's all hosted by bit or literally by your website, essentially, which.

Steve (27:30)
Yeah, Bitbucket is just for the source code. It runs the build script, which deploys to the Dominion internal IT system. we do our own hosting here. We have a hosting team for security and things like that. they have their own servers that they run. So it's deployed from the.

Craig (27:51)
But Bitbucket

handles version control essentially. Yeah.

Steve (27:56)
Yes.

David Moses (27:56)
So

I would, yeah, I'm surprised if instead of copy paste to an email, at least whoever hosts your email, you should probably be able to silo just draft creation. Like create email.

but create draft only, and then at least they can, you the person who's actually sending it can go to their draft folder and open it and send it, where the system can't send it, but it can actually create it. You should be able to silo those permissions. I mean, certainly with Microsoft you can, I assume with Google, you know, with Gmail you can.

Steve (28:29)
Yeah

Yeah, the ⁓ problem is not so much technological. It's more a question of just time. I've only been at this a month and that's kind of sending emails is kind of phase two. But yeah, I have.

David Moses (28:45)
I got you.

Yeah, I thought the issue was permissions,

like getting them, getting the IT team to be okay with that.

Craig (28:54)
It's that too. It's that.

Steve (28:56)
It's a little of both. I

Craig (28:57)
Yeah.

Steve (28:58)
have to jump through the security hoops to get the permissions, but it's not a difficult process.

Craig (29:03)
All right, Steve, keep going.

Steve (29:04)
So if there were anything missing, let me just uncheck one of these items. I'll do Save. You can see here that there's an export requirements card, we call it down here. And so you could open that, and it'll copy it to my ⁓ clipboard. And let me pull up a notepad here. ⁓

There we go. And so you can see there's like an example email that that's what it would look like. You hey, I'm reviewing your deal at such and such an address. I need this specific information to complete my underwriting. Please, Yeah. So, yeah, eventually that'll be automated. It'll be fully automated. The underwriter will have the opportunity to edit the email and then click send. It'll be great. Let me just.

Craig (29:43)
McTester face is that Irish? That's Irish name, right? That's cool.

Steve, ⁓

quickly, I'm sorry, on the SMS feature, will you use like a voiceover IP phone SMS? Will that come from like a call rail or a ring central number or are you using like iMessage? What are you doing there?

Steve (30:18)
Yeah, for SMS I was gonna use Twilio, I think it's called, is a provider that it's pretty automatic API enabled and you can just send text messages willy-nilly.

Craig (30:30)
I don't know about Willy nilly. I think that you have to go through like, like the fed regs, ⁓ for SMS functionality. I found that to be a bit of a pain in the ass. And I was like, I'm just going to go through my iMessage at this point for any text messaging from my bot, which is kind of clue G but, ⁓ yeah, I think I'm more hope we're getting ready to implement ring central here, Dave. And I think that's going to be a much more robust product.

David Moses (30:38)
That's right.

Craig (30:58)
call rail which we use now.

David Moses (31:01)
My experience is that if you're gonna send mass text messages, get a Twilio number. If you're text message anybody who's not already in your ecosystem, I would get a Twilio number. Because if it gets spammed, okay, so we're just not gonna use that number, we'll just get a different number. But yeah, on the production side, if it's trusted, people you trust,

Craig (31:14)
Mm-hmm.

David Moses (31:22)
that you're already doing business with, doing it right through CallRail API or we use Dialpad, it sounds like guys are going to RingCentral, which is expensive but a good product.

Craig (31:34)
Yeah. Sorry man, keep going.

Steve (31:37)
That's all right. Anyway, just behind the scenes, SMS is grayed out because I don't have that hooked up yet. It'll be a future enhancement. And there's also going to be a shared spreadsheet option here. So let's say the borrower is doing a portfolio of 20 properties and I'm missing a lot of details. I don't want to send individual emails. I don't want to send text messages for that. I'll just share a Google Sheet and say, you know,

Hey, Barra, here's the 20 things I'm missing. Please fill this spreadsheet in and it'll automatically come back to me because it's a shared sheet.

So next would be budget. budget used to be a pain in the neck. Now you just one click and it loads it from Salesforce for us. We already have that information typed in. So I know what their purchase price is. I know how much rehab they need. And I can add and subtract different items if I wanted to here. This is not individual line item budgets. This is just overall. What's your purchase price?

Is there an assignment fee? Are there seller credits? All of that sort of thing you're going to add to your budget tab here.

And as you can see, as I'm marching down the field here, the chips above each button are turning green to let me know, hey, you're done with this phase, you're done with this step, and you just keep going. Halfway through the book report, exactly. Next up is approval.

Craig (32:54)
You're halfway through your book report, sir. Keep going.

Steve (33:00)
We use an approval process that's we call it RTL3. And that includes a narrative where we describe what is it that went into the approval for this borrower? How did we decide what we would quote on this deal for this borrower? And to do that, we look at their experience and we look at how many deals they've done of each different size, like dollars per square foot on rehabs.

Have they done ground up construction and all that sort of stuff. And we summarize all of that into just a couple of sentences and you can see here what the sentences look like. But to get that, it's kind of like sausage. How do you make the sausage to come up with this one sentence? And there's a lot of things that can feed into that. A lot of things that underwriters look at in terms of figuring out how much we can lend.

And depending on the deal type and how much dollars per square foot and things like that, that can change what you see up here and what the approval is. So I don't need to go through that probably piece by piece, but the important one is the green card here, which is approval terms. And that says, you know, this borrower is approved for a hundred percent acquisition. So in other words, I can pay a hundred percent of the purchase price of this property.

That's one of the things Dominion does for ⁓ experienced borrowers on a deal that's kind of in their wheelhouse, very simple and straightforward. can fund 100 % of LTC. So ⁓ next one would be LTA RV. So what's the loan to value that we can quote? What would be this borrower's interest rate, their loan term, some of their fees, all that goes into there.

And you can see what that looks like, that approval process looks like in what we call the underwriting matrix. And what this shows is the borrower's experience for each different type of deal. And we kind of line it up and figure out where they belong. So this deal that I'm looking at is a single family rehab. So that goes into this first block. And then within that, there's different price tiers. ⁓

Does the borrower have any experience in $100 to $200 per square foot single family rehabs? No, they don't.

Craig (35:31)
Yeah, one of our one of our overlays here, Dave, is price per square foot for rehab. Like, obviously, if you're someone who has experience doing 20 to 60 dollar a foot rehabs, that's a hell of a lot different than bringing us a deal where you have no experience doing a luxury two hundred and fifty dollars square foot plus rehab. And so we look at that differently. If a guy brings us a deal, even though he's experienced, do you really have something that you've gotten to the finish line that's going to be that?

you know, that detailed and that much rehab. And so we would probably back off on our max LTC, LTV. And in those cases where a person, it's not as if we wouldn't fund the deal. We're just not going to fund it at 100 % LTC.

David Moses (36:12)
What's

the timeframe you're looking back in terms of experience and how are you building that? I mean, I think the older you go, the tougher it is to find that data, I assume. I don't know.

Craig (36:25)
So three years ⁓ for experience, look back, and you could probably answer the second part better, Steve. So how far back do we go? Three years. And then what was the other part of the question, Dave?

Dave.

David Moses (36:39)
the other part of the question, you know, I was just, it was how long do you look back, but then, you know, is it more difficult to go back further and get more information? You know, a lot of people, I mean, like just using me as an example, I'll follow where I think, you know, where I think the opportunities are.

So I may spend a few years in one type of deal and then kind of shift to a different type of deal if that's where I see the opportunity. So like ground up construction, I've a dozen houses, but the last house I built was in 2018. So it's a while ago. That kind of thing, because you just kind of shift from.

Craig (37:03)
pivot.

David Moses (37:17)
one strategy to the next as you see the market opportunities present themselves.

Craig (37:22)
Yeah, I think being a private lender, can look at, you know, the story and kind of write the rules as we go. But I think generally speaking, these RTL three underwriting guidelines, the matrix is something that we've employed now for probably a year and a half. And ⁓ it's essentially Jack's risk overlay for.

you know, for guys who have the requisite experience that we would want to fund deals at a higher LTC, LTV, right? Like I don't even, you know, in your case where I was a builder 10 years ago, I don't know that we would give you that credit. Like, Hey, great story, but it's been 10 years and Lord knows the labor and material markets have changed. You know, permitting has changed in places. So go out and do a couple more of those and then we'll start to consider you a builder again. Right.

David Moses (38:11)
Yeah, that

makes sense.

Craig (38:13)
Keep going, Steve.

Steve (38:15)
Sure. anyway, the various pieces here all funnel into this one sentence, is, here's what we're quoting on this deal. Here's what the borrower's approved for. Doesn't mean it's written in stone. We make modifications to it all the time. But that sort of gives us a starting point to quote from. And there's our narrative. So I'll just save and close that. I can also get a sample quote at that point.

And now we get to comps. Comp's in value. This is where the underwriter spends most of their time. So ⁓ comps, yeah, where we earn our money, exactly. So just to give you an idea, like comps, I mean, we've all pulled comps, Anybody involved in real estate knows what a comp is. Comparable property to this one. ⁓ But it used to be you were always looking at one source of comps at any given moment.

David Moses (38:48)
worth air and their money.

Steve (39:12)
go to one website like House Canary or Zillow or Redfin or any other MLS, exactly. But you're only looking at one source at a time. You'd open one website and of go through and search and look for different features based on what kind of deal you're looking at.

Craig (39:17)
Maybe you have access to the MLS, right?

Steve (39:34)
But the problem with that is that no single source is really amazing at everything. And you have to change source depending on what city you're in, what state you're in. Sometimes you're quoting a deal like in a non-disclosure state like Texas, there might be no comps, right? There might be no way to find you. to Zillow and everything's in the sold. If you do a sold search on Zillow, everything's yellow, like just.

no values at all, just dash dash where they're supposed to be a sold price. And that can be very irritating, but we have different vendors in different places that are better at those than others. And one of the sort of the driving forces when I created this app, I'm like, I am going to solve this problem, right? I don't want to open 20 different websites to track down comps anymore. I got to fix this.

And what I came up with was the idea of I'm going to do a comps marketplace. And instead of having just one, like, okay, now let's search the MLS. Now let's search how skin area or whatever. I'm just going to give the underwriter the ability to check as many providers as they want and search them all at once. And I'm going to consolidate all the results into one Google map with pins.

And that's what you're looking at here. That's what I came up with. So we have our providers across the top. And in this case, it's ⁓ RESTB, House Canary, Data Tree. And these are dynamic. I can add and subtract depending on the deals, on each individual deal. ⁓

I can also add different styles of comps. like these are sold comps, these three that you see here, but for multifamily, I'm probably going to want to add rental comps. And for vacation properties, I might want to add Airbnb comps or, know, ⁓ Verbo or whatever different types of comps I want to get from different sources. And they're all available. So it's really easy now. Everything is just effortless with AI development.

Anything with an API, you can go out and pull data. I can pull data from public records. I can pull data from my own deal is another option. If I'm doing a multi-family prop, ⁓ multi-property portfolio, I can say, what comps did I use on that other property? And just add them automatically. So ⁓ you'll type in your search criteria. You can pick how far out do you want to go and how old do you want to go.

And when you're done, you hit search and it will go out and find all these comps in seconds. You can time me on this, maybe like 10, 15 seconds, something like that. Boom. So now you can see I've pulled comps from all three sources all at once.

David Moses (42:20)
wow.

Steve (42:28)
And then I can zoom in and see the different colors. So if you look at the key underneath here, you'll see how skinnery properties are blue, data tree only properties are yellow, REST B comps are green, and then any comp that came back from multiple vendors is going to be purple. So those are kind of hybrid.

Craig (42:51)
Dave, your head's spinning. I can tell.

David Moses (42:52)
I love it. This

is great. you, like, when it's grabbing these as comps, like, is the underwriter choosing the criteria to determine whether, like, okay, this fits the basic criteria of a comp, but is this really a comp? Who's making that decision and how is it made?

Craig (43:18)
So, so yeah, I'm looking for all houses that were, you know, one month, less than one mile away, 12 months. and ⁓ yeah. But then is it also looking for comps that are closest to the subject property given all of the other steps prior? So it's not going to pull up a 3000 square foot house if I've got 1100 square foot subject property.

David Moses (43:39)
No, sure.

But you're not going to look a mile away in 12 months in City of Baltimore. A mile is a long way. As I know, Detroit, it's a long way.

Steve (43:41)
Well...

Right, yeah, you can narrow this.

Craig (43:50)
the universe.

Steve (43:54)
Yeah, this one, these are not comps that have been selected yet, right? So I'm giving the underwriter, here's the shotgun blast. Just show every possible comp that I can find in the system and pull them up here on the map. But then the underwriter is going to click through individually. The underwriter is going to go one at a time and probably start with the closest ones, especially if you're in a big city like Baltimore or Detroit.

Craig (44:03)
Cool, yeah.

Steve (44:23)
you're going to start with the closest ones, click each one, and just eyeball it and say, hey, does this look like it's going to be a good comp? And they can click through the photos. Some of our vendors supply photos, some do not. So like that one had photos. House Canary's properties in blue, they usually do not have photos. So I'll have to go out and retrieve those manually. The green ones generally will come with photos. So let's say I click through here.

Craig (44:50)
Such a great, such

a great little pop up there. That's a great little, yeah.

David Moses (44:52)
Yeah, who's

serving you? Which ones are serving you the photos?

Steve (44:56)
Rest B is serving photos and ⁓ then after that for House Canary and Data Tree they don't supply photos but I have a technique for retrieving them. So ⁓ we're going out and scraping our own photos when necessary.

David Moses (45:07)
Gotcha.

Gotcha, gotcha. And those are MLS photos. That's not just a street view photo. That's a listing photo.

Steve (45:14)
So let's say I look at this one.

That's right. Yeah.

Yeah, and this is a miracle, right? mean, it used to be you'd have to open up five different websites and comb through the photos. Now you have them just at your fingertips. So let's... ⁓

David Moses (45:23)
Very cool.

Do you have AI doing any kind of analysis

on the photos themselves?

Steve (45:36)
Not yet. ⁓ REST B is an AI-enabled partner, and they're actually analyzing photos with AI. And they hand us back metadata about each photo. Like, hey, this is a kitchen, and it's got quartz countertops and stainless appliances, which is pretty neat. ⁓

David Moses (45:50)
sweet, okay. So you can use their already

processed data to sift through. That's awesome.

Steve (45:55)
Yes. We're,

I'm not doing anything with that yet, but that's definitely on the horizon. We will be narrowing that down in the future.

Craig (46:06)
for plotting the comps themselves. Is that a mix? That's that's a Google API thing, right? To plot the little to plot the points. Okay, great.

Steve (46:16)
Yes. Yes. Yeah. I'm

handing Google maps an array of, here's my comp locations addresses. Please select them and give them a pin. And I tell it what color the pin is and what the pop-up and it just pops up.

Craig (46:32)
that's so cool. I love that. I love that for a very specific reason that I'll share later. Keep going.

Steve (46:35)
Very slick.

⁓

So the underwriter picks their comps. again, they could still open up other websites. They can still go out and talk to the borrower and look at Google Street View and everything about it. I have some automatic.

searches that they can do over here. So we can open up, run a Google search based on this exact address, and it'll show me what the Google results are. I can open the Zillow link. It'll go straight to Zillow. And I can also go out and try to scrape pics off the internet if I don't have pics already, but I don't need them here because I got pics already. So when I'm done choosing my comps or the underwriter is done choosing their comps, they'll save changes.

Just going to confirm three comps.

And that's it. So now I got my comps. ⁓ Just for backstory, it's not usually that fast, right? The underwriter is going to spend quite a bit of time crafting the perfect... Yeah, a lot faster than it was.

David Moses (47:38)
sure.

Craig (47:41)
It's a lot faster than it used to be.

David Moses (47:45)
And the underwriter is, ⁓

you know, they're getting that choice. Are they making any sort of adjustments to what they believe the value is based on those, like, okay, I've selected these three cops. I got you, sorry, we're getting to value. apologize.

Steve (47:53)
Yeah.

We're getting, yeah, you read my mind. It's

exactly the natural progression is ⁓ now we're going to do the value. So then we open up the value. This is going to show me the photos. Here's a street view of my subject. If I want to see what the subject looks like, I can also upload photos. Like if the borrower sends me photos, know, hey, it's a.

it's an auction property and I went in toward the house and here's the photos I took or something like that. I can drag those in and add those to the deal. But up here we have photos of the actual ⁓ comp that's selected and we can go through and click them. This innovation, having the photos right on the adjustments page is huge because

It used to be you had to go to third party websites and you'd have, again, multiple monitors and multiple windows open and trying to see what you're adjusting for. Now it's just right on the page. You can click through and see the photos. You can make them big. can do any kind of...

Craig (49:04)
Steve,

this point, are the photos all being stored locally? Are you storing them in some sort of super base at this point? Like, there's a lot of assets being collected here for an underwrite. I'm just wondering like, and storage is cheap, frankly, but I'm just wondering like, at what point does everything get stored?

Steve (49:16)
Yeah, I-

As soon as you pick it as a comp, I'm downloading it. It's now cached on my server and I'm going to save those photos for the duration, which is till closing, so like a month maybe, something like that. ⁓ But yeah, I'm actually downloading them and they're live from my server.

Craig (49:27)
cool.

Mm-hmm.

David Moses (49:42)
Why

not save them through the life of the loan for an RTL? Just out of curiosity, if it goes bad, then wouldn't you want that stuff to kind of do a post-mortem on?

Steve (49:52)
Yeah, yeah, that's, we actually are in the background. We are caching them to a longer term storage. have like a, I would call it warm storage. exactly. Yeah. Well, we have three levels, right? So there's hot, warm, cold. so I'm downloading my server for hot storage. Instantly I can get it through loan closing.

David Moses (50:00)
like, I gotcha. Glacier storage. no, warmth.

Steve (50:19)
And then later on, if you want to come back and revisit it, it'll be in warm storage exactly for that purpose. We can use them for draws. We can get a second look when we're doing draw requests later or, like you said, a postmortem if something goes wrong. Yeah, we have backups.

Craig (50:34)
Dave, I've created a site where, and it was just really on a lark that we could use for marketing and sort of customer experience where it's like a story engine for your deal. And so we know what the purchase price is. We've, we've got all the data on the house. And so what if we created a unique website for you that kind of tells the story of your rehab throughout the process from start to finish, cause we're going to get all the progress photos as well.

And we upload those to this really beautiful experience website. And then you have this living document that catalogs the story of the start to finish of your rehab deal that you can share with your equity partners later. And you can share with your

David Moses (51:17)
But I want to be careful with that. They may

share that with banks or other lenders.

Craig (51:20)
Yeah, that's true, but we're

hoping that like, you you give them that little that little tidbit there and they're like, you know, Dominion cares, you know, so keep going.

David Moses (51:24)
It's a value. Yeah, no, that's awesome. That's

actually really cool. And this is all the borrower sees which comps you picked and why. If they want to say, I think you missed this comp, or I don't think that's a comp for this reason. Is there any back and forth on that?

Craig (51:43)
We'll keep

Steve (51:43)
Yes.

Craig (51:44)
keep going Steve, show him the report that the the that the borrower actually gets.

Steve (51:49)
Yeah, you want me to jump ahead to that now?

Craig (51:52)
Yeah, sure. Keep going. Take your time.

David Moses (51:52)
you can keep going, it's fine.

Steve (51:54)
We'll

get to, I'll show you the report at the end. But yeah, there is a report the borrower can see.

David Moses (51:56)
My memory lasts at least

15 minutes, so we're good. After that, it's like, what's my name?

Craig (51:59)
All right.

Steve (52:01)
Yeah.

So after you've selected your comps, you have your photos up there, you're going to do adjustments, of course, to sort of narrow down your kind of triangulating the value of your subject. And this table is not using AI, it's just using math to say, anything that kind of is standing out of the ordinary compared to your subject, it's either highlighted red or green.

telling the underwriter, hey, this might be something you want to adjust for. That's all. So in this case, I picked three comps that are somewhat bigger than my subject. these are highlighted red. And they're saying, hey, underwriter, you might want to put a negative adjustment here. So a 1,400 square foot comp is probably going to be worth a little bit more than.

and 1,100 square foot subjects. So it's saying you might want to subtract some value. And it has a recommended value that we would use. So you can just hit Apply, or you can type in whatever value you want. I'll just hit Apply for all three of these. And then anything that's green, it's saying, hey, you might want to put a positive adjustment. So these are much smaller lots. And our subject has a much bigger lot. So you might want to add a little bit of value here.

Craig (53:11)
you

Steve (53:21)
Year-builds are a little bit older. So you might want to add a little bit of value for that and then know bathrooms, of course My borrower is going to be converting a half bath to a full bath. So you probably want to add some value to that Let's let's say you think that's worth, I don't know 7,500 in this marketplace and I'll go ahead and type that in

David Moses (53:41)
So

I always wonder, and I'm sure AI can solve this maybe now or in the future, I always wonder, these appraisers use these kind of round numbers and kind of guesses at how they're going to adjust for certain things in certain markets. But I wonder how those meet up with reality.

I suspect that some of them are pretty spot on and others are like, wow, that's, you I wouldn't expect to have that kind of adjustment if I were selling the house and it were, you know, bigger, smaller, you know, whatever the situation is. Just a thought.

Steve (54:25)
Yeah, yeah, I don't know if there's really been any study on what the appraisers use, but if anybody who's worth their salt should be basing it on the individual marketplace, right? You should be looking at a wide range of comps to get a feel for, hey, how much is a bathroom worth? And that changes over time and it also changes over location.

That's how we do it here at Dominion. We would be surveying the exact marketplace on the exact date that we're doing the underwriting to figure out how much is that extra half a bath worth or whatever it is we're adjusting for. Sometimes you run into weird things like, hey, I'm in Baltimore and this guy has a pool. Well, right now, how often do you see that? Not very often.

David Moses (55:13)
you

Craig (55:18)
not exactly a

selling feature in November.

Steve (55:21)
Yeah, exactly. That might even be a negative, right? You might be trying to subtract for that in this case. And in those cases, you know what? You just kind of have to do the best guess, educated guess you can do. But ⁓ something like just a half bath, that should be pretty easy to research.

Craig (55:40)
All right, we're 59 minutes in. Let's land the plane.

Steve (55:43)
Okay, so we're going to do our adjustments. We'll usually add some commentary and the underwriter will put in a final value. Let's say I think it's worth, I don't know, $300,000. And I'll just save and close. As you can see, we got it 100 % green. All of our chips are green. At this point, I'll just do a finalize. And that's going to lock my property adjustments and spit out some loan quotes.

So I'll give it some deal notes and it gives me the outputs of, know, hey, here's how much we can lend. Here's our interest rate and our, you know, loan term and all that kind of thing. And I can also produce a final PDF. You were asking about the PDF days. So let me create that. I have a few different ways to do that, but ⁓ it used to be that what we would do is go out and

take our final underwriting report, like a list of comps and the adjustments we made and all of that sort of thing. And we would go to a third party vendor and ⁓ paste in our report and they would create the PDF for us with all of the comp photos and things like that. And usually that was house canary. They have a little commentary field you can paste in your underwriting report. But in this case, ⁓ because everything's AI enabled now and ⁓ really easy to do, I'm generating my own.

PDFs now. So I've already downloaded the comp photos and I just need to ⁓ click the button. It's a little bit slow today, so I should have added it. I get a green note that says, hey, we've added the underwriting report. And here it is. It gets added to my list of deal files. So I'll open that up.

And you can see what it looks like. Ta-da. So here's my subject. I got a street view photo right up at the top. This comes from Google street view. And instead of, you know, some kind of third party branding, this is the official Dominion branding. So we have our logo on there and it tells me who created it and what the date was and who the borrower is. You very customized, highly customized just for us.

Okay, so next thing we're going to do is we see some sort of summary of the property in the deal and kind of what kind of property is it and the basic stats. Next up is the underwriting report. So this will have my adjustments and what ⁓ sort of went into the value that I came up with. as you can see, it lists the addresses and a little bit of credit info up top.

⁓ And then we have here's a summary of comparable sales. So it's showing me a pin where my subject is and a pin for each comp and some comp data. And then down here is kind of the beautiful summarization, I guess, where we actually get the photos of the comps and ⁓ kind of a breakdown of everything and what the adjustments were.

So very cool stuff. After that, have some internal research. show this is an AI-generated summary of what the borrower history is and their experience profile. Here, this one is couldn't find the name testing McTesterface, so it picked some random investor that isn't the one we wanted. But you get the idea. So that's what it looks like.

Craig (59:20)
Steve, what do you think ⁓ the cost? So here's my thoughts. If you're a listener right now and you're watching, you're saying, well, this is is crazy stuff. It's huge. But like I have never written a line of code in my life. I think I'm fairly technical, but certainly not to Steve's level. But I feel like I'm developing similar platforms just based on my.

David Moses (59:20)
Really cool.

Craig (59:45)
seven months of working with these tools. So if I'm a listener to the show right now, I'm watching this and I'm just an investor. I'm a borrower.

What's my takeaway? Well, my takeaway is, man, if I had the same access to similar APIs, how much better offer could I make on a house quickly? And how could I track my deals and sort of how could I track motivation in my market? And I think at least with the API access that I have to like say, elementics and house canary and some of these others, you can, any person could develop a report, you know, a similar piece of software like this, not for

not for underwriting, but really for trying to figure out what they're going to offer on deals, what their market looks like. And so the quick question, if I'm a regular Joe and I want to gather this sort of data, what am I paying for that every month, Steve? Like, give me like a, am I paying like for the Claude code access, like maybe...

a hundred bucks a month, 200 bucks a month, and then some of the API access is like, what am I looking at there? If I'm a guy and I want to develop something similar to this.

Steve (1:00:52)
Well, there's a couple of ways to look at it. We have subscriptions that are monthly, and we're buying thousands of reports. that's.

Craig (1:00:59)
And we're

buying nationwide data rather than just some of our market data, right?

Steve (1:01:02)
Exactly.

So our subscriptions are going to be very, very expensive. We're spending tons of money on massive subscriptions that probably is not what an individual investor would need. On an incremental basis though, everything we just looked at is dollars, like maybe three, four, five dollars worth of API calls per.

Maybe that includes preferential preferences. But the marketplace is kind of, in my opinion, it's heading in this direction, just like everything else in SaaS these days. It used to be like everybody was pricing everything per seat. But now it's kind of switching over to price per work, right? Price per API call or price per address that you pull data on, stuff like that.

I think that on average you're probably looking, I might have made, to do everything we just talked about, I might have made 10 API calls that cost money, and I bet you they averaged about a dollar each.

Craig (1:02:04)
So Jack and I, you know, we pride ourselves on the real estate, the real investor radio podcast. It's not, it's unabashedly, you know, for people who are advanced investors and a lot of guys that we talked to are doing 25, 50, a hundred plus deals a year. And so to have a dashboard like this, that literally gave me market data down to the zip code of areas that I know I liked investing in. I think it

be invaluable for guys like that. And in this day and age to have that sort of sizable real estate operation and not have access to stuff like this at your fingertips, I think it puts you way behind the eight ball if you're trying to develop it 12 months from now. Dave, any questions from you? Any thoughts, comments?

David Moses (1:02:51)
I think it's

super interesting, kind of the way you consolidated it, the way you did it without writing code. You're using the tools available to make the code. I think for somebody who is thinking about, maybe they are a small lender or maybe they are, maybe they're just, hey, I wanna underwrite my deal at least as well as Dominion is. They're putting in debt.

I'm the one putting in the equity. I'm the one who's putting in the work. I should at least be as good as they are in underwriting my own deal. So I wonder if it's something that parts of it, you know,

point people in the right direction in terms of, hey, if you grab this and this, you're going to be a good chunk of the way there. Lestam Indian plans on just saying, here's our info, pay us money, and underwrite your own deal through our system. And we're still going to underwrite it, but you can at least do it yourself and use it.

Craig (1:03:47)
You could do it. Well,

we'll think about this, Dave. mean, in this day and age when transactions are, you know, great transactions are a needle in a haystack and you're and let's say you're in a market where you're competing against Dominion properties or someone of that size.

You better be fast. You better you better be able to make a fast offer to a seller who's ready. You know, because you ain't going to be the only one sitting across the table from them. And I guarantee you, the guys who have these tools are going to be able to make much faster offers and probably be a little bit more spot on. So if I've got if I'm a guy who does 50 deals a year and I'm looking to scale to 100 and I don't have this kind of market data.

Yeah, think you're just going to lose the speed of offer, speed of service to that motivated seller.

David Moses (1:04:37)
Yeah, I think that's right. I think that's correct.

Craig (1:04:40)
So Steve, add on features, like stuff that you're excited about. Like I was talking to Jack the other day and I think Dave, you can speak to this. Like I think we live in a day and age where it's like really important to get Rev1 up and running as fast as possible.

like put something out there that like is a little battle tested and then really test it out in the wild and then just feature the shit out of that thing with Claude code and say, hey, I, you know, these are the 10 features I'd like to add over the next day. And it's just start spitting out features. And so given that I love the fact that like, you know, you brought something that's really hardened and works really well as like Rev1 it's, you know.

To call it Rev1 is kind of shameful because it's so tight. But what's the add-on module right now? What's that killer feature that you think is missing that you're excited about?

Steve (1:05:29)
So the next big thing is workflow. So the underwriter is going to go through here. the net result, the end result of this underwriting that we produce is we still end up sending a plain old email. We write what's called the investor committee memo. So we write up a summary of the deal. We send it to Jack. Jack approves the deal or modifies the deal or whatever.

And then it goes to the loan officer with a loan quote. And they quote it. And that's great. That's all done in old-fashioned email right now. And Jack wants the ability to do all of that on his phone. We need a phone app. And so when the underwriter is done, why are we producing PDFs at all anymore? Why are we doing any old-fashioned email? Instead, we're going to click a button in version 2. This is coming soon. We're going to click a button.

it's gonna go to Jack's queue for review. Jack will get a little alert on his phone, it'll ring or something and he'll flip it up and just scroll through real quick. Boom, here's the loan quote, here's the underwriting, here's the comps. He'll be able to see the Google Street View of every comp we picked and the subject and.

see all the graphs and charts, see all the AI work, anything that needs to be done, any research in the background, and then just instantly click yes, approve it or don't approve it or modify it like this, and then boom, send it to the yellow. That's the next big leap in this process.

Craig (1:06:54)
you

David Moses (1:06:57)
Yeah, think that's right. think what Craig was saying in terms of just getting something out there into the wild, whatever your business is, I think that that's...

Craig (1:06:59)
So love that. Go ahead, Dave.

David Moses (1:07:10)
really, really sound advice because you'll learn one thing that you expect to learn and one thing that you maybe don't expect to learn. The thing you expect to learn are, know, whoever the users are of this, they're gonna help you prioritize the features you're thinking in your head might be good and come up with new features that might be good for it. But the other is a gauge of buy-in and I think that the things that I've deployed in my business

have really helped me to see who's buying into where we're going and who's really gonna be left at the station while the train leaves. And that's something that may be unexpected, but I think that putting these things out there before they have every bell and every whistle, before they have even every bug fix, it's important especially for a non-coder.

you know, to get it out there. Because debugging and adding features is actually easy and fast. Knowing what to do, knowing what's going to have real impact, and then gauging the buy-in, you know, that stuff is going to be, I think it's significantly more valuable.

Craig (1:08:20)
So I have a question for both of you. So let's say I develop a app in OpenClaw.

We can't really run open claw in the wild here in production at dominion for a multitude of reasons that we're just not willing to like get, go through those hurdles yet, but we can run Claude code and it's being deployed almost to every employee in the company. Can't imagine what the token cost is going to be in just a month or so as these people who have no clue what they're doing, just start banging on the API. So question.

I develop, let's say I develop an app. I'll give you an idea. So I've, I'm going to focus solely on small to midsize lenders and building the Dominion lender finance business where we lend to lenders.

And let's say I have a really great list that I pulled down. I've pulled that into my CRM that I've made. I've enhanced it with elementics data. And I really have an unbelievable screenshot of what each of these lenders are doing, what their competitors are doing in their neighborhood, and who the borrowers are in their neighborhood. I can pull that immediately on every lender that I speak with.

The question is, so now I'm ready to go enterprise with this thing and port it over to something that's in production, but it still has agentic workflow in it. And that agent has sort of been tuned in the build, let's say. it's a smart agent. How do I then deploy all of that agentic ability?

you know, to the production app. Like, am I just giving it the same API that I used? Am I going to give it like some other LLM that's cheaper? You know, so Steve, speak to that. Like, you're now in production with this thing, but it has an agentic flow to it, and it's an agentic information. Like, are we pulling an API from this server, this app that's now on the server? Is that?

Essentially what you're doing just pretty simple stuff

Steve (1:10:23)
Yeah, yeah, great, great question. So I just did exactly what you're describing. Everything you just described, I just did it. And it took an hour. It was easy. So let's say you sort of organically created this report over time. You've been interacting with your Claude bot or.

Craig (1:10:38)
Mm-hmm.

Steve (1:10:43)
Cloud, web, or any AI to get to this point where you have this amazing report that you like and sort of a set of queries that got you there. Jack did something very similar to that and handed me, hey, Steve, can you add this report to your underwriting report? And it was this really complex, detailed, agentic style.

Craig (1:10:44)
Sure.

Mm-hmm.

Steve (1:11:07)
hunt

for data about the borrower and all their previous mortgages and previous fix and flip deals and things like that. He said, can you get this into the app? And I said, sure. What was your prompt? And he went over and copied and pasted his conversation, just literally his conversation with the AI, and sent it to me and said, here you go. And what I did was I took his output report.

and his conversation with AI. And I pasted that into Claude code. I said, Claude, here's this report that I would like to produce. Here's the prompt that created the report. And here's the APIs it used. And here's the access to these services that Jack was using to get his report. Can you reproduce this?

Craig (1:11:33)
Mm-hmm.

David Moses (1:11:37)
First engineer this.

Steve (1:11:55)
And guess what? course, Claude, is, yeah, of course I can. So it recreated the prompt and recreated the report based on what I had presented. And it then incorporated it into my app. And the whole thing was, I say an hour, was really a lot less than an hour.

Craig (1:12:11)
Yeah, but

so now it's incorporated into your app and it did that first one as a test, but now it's in production. What LLM is it still going to call the same LLM that you use for the test or do you crank it down to something that would be less expensive?

Steve (1:12:32)
Yeah, you get your choice. So it'll ask you, which model do you want to use to make this call with? And I have API keys for every major AI model out there, and I can just pick. And one of the things I did in that process was I asked Claude, well, which one do you recommend? And it recommended Claude.

Sonnet 4.6, which is a little less expensive than the flagship. But I also experimented with a few others and I picked the one that most, you know, produced the most natural result for what I was aiming for. Just reading through the reports, I picked the best one. And then I said, okay, that one, and I gave it an API key and now it produces it for any borrower on any report at any time.

Craig (1:13:15)
Makes sense. Got it. Dave, anything there?

David Moses (1:13:18)
Yeah, think that's, I think.

What I think people really will benefit from is understanding that all of these agentic workflows are essentially, they're all built with the same formula. On one side you have the context, right? Here's all my information, I'm grabbing it from APIs, I'm grabbing it from internal data, where am I getting the information that the LLM is gonna look at? So that's the context, and then we prep that with a prompt and a system message. So the prompt is gonna say,

here's all my context, hopefully if we do it right, it puts that context in line, so it says this piece of data I'm gonna use to determine this and then it plops in that data and then so on and so forth. So here's my context, here's my prompt, which is gonna be what am I doing with this context, right? And then a system message that basically just gives the LLM or the agent a persona, like here's

how you feel, here's how you, kind of what you're trying, who you are kind of thing. And then a response, right? So if you tweak, yeah, exactly. that's essentially it. It's context, prompt, and if you can just tweak context and prompt, you can get some very dumb.

Craig (1:14:22)
What's the output?

David Moses (1:14:33)
l l s to do some really smart things if you give it enough context and give it specific enough instructions ⁓ and then you can you can learn

Craig (1:14:38)
Yeah.

David Moses (1:14:40)
you can learn from the most, from the smartest LLMs. I often will start it on OPUS and then have it do what it does and then have OPUS tell me, okay, based on the reason you went through to give me this output, give me a much more detailed prompt that would help a lower cost LLM to come up with the same result.

Craig (1:14:59)
very cool.

David Moses (1:15:00)
Yeah,

right, so it'll just give me, but that's essentially it. And I think everybody in every business everywhere really has to understand that completely. If I can create a workflow where I just have context and prompt, I can do some really incredible things.

Craig (1:15:18)
Steve, there's so many questions I could ask you about the UI because for me, it all comes down to the experience, the user experience. just, I don't know why I care about it so much. just, I can't put out like a website that I think is like shitty from a user experience standpoint. So I spent a lot of time doing that.

And I was looking at Rachel's program that she's doing yesterday and yours. And what I'm really most intrigued by is the tables that were generated in your report and sort of the really just simple yet elegant layout.

I, we've gone too long to get into all of that, offline, would love to show you like a couple of things that I'm building and like wonder like, well, why can't, why doesn't my AI generate like really good looking, easy to read tables? You know, it gives me something decent, but it doesn't always get it right. And I'm really in that stage right now where I want to get it right.

So, um, love to talk to you about that offline and just sort of like, how do you, how do you really tweak a UI guys? Yesterday I developed Dave, I developed this thing yesterday where I can add an inspector layer to any page that I, that I develop. So let's say I develop a page and it's got a header, a footer, all these different containers, know, colors, weights, everything. Why I do a page anatomy. I haven't, I have a bot that gives me an entire page anatomy and literally every single thing on the page gets

a name, right? It's just a file. It's just a name. And then I can do the inspector, which is an overlay. And I say, show me all the layers. And it does a little overlay with the layers. Show me all the elements. And so now I can literally click on anything on that page and talk back to Claude in a way where it knows exactly what I'm talking about. Not like, hey, can you change this header? Hey, can you move this over a little bit? Hey, can you make that button?

I can show it exactly where I am on the page and then give it a prompt to go right to that area, which took me about an hour and half to do that yesterday. I'll show it on the next podcast. It's really cool. Yeah, to tweak a UI, that's all I was doing yesterday. And I would say I probably got 3X done in tweaking a UI yesterday just because of this little tool that I built.

David Moses (1:17:30)
I mean, it's cool. I don't trust myself on UI. What I basically do is I say, you know, hey, here's a website that I think really kind of utilizes the functionality that I'm going for here and then just have it kind of review it and try to mimic it as best you can. No, no. But it's a much better starting point for me than...

Craig (1:17:40)
Yeah.

Yeah, yeah.

Doesn't always get it right. It's the tweaking part that takes all the time, Yeah, it

is. Well, Steve, thank you so much for taking the time. I think it's just the reason to have you on was one to show off incredible work. it's just a testament. It's just a testament to, like,

I started this six months ago, I started this two months ago, and I now have a tool that is comprehensive and does underwriting in seconds rather than underwriting in a day. I mean, if there's not a better testament to getting started with these tools, I don't know what is, and I just can't thank you enough for taking the time to show us.

Steve (1:18:28)
It's been an honor. I'm very excited to meet you, David, as well. I'm a fan of the podcast, too. I listened to the first episode and I thought it was really terrific. Learned a lot myself. So thanks for having me on.

Craig (1:18:40)
Well, we have so much more coming. Dave, are you working on anything that you wanted to show off on the podcast? By the way, we kind of have this thing where like every podcast, we're all supposed to come up with something that was cool that we built during the time off. Dave, anything that we can show on the next one, maybe do a mini podcast and show what you're building on the next one?

David Moses (1:19:00)
Yeah,

so I'm building, I remember like last time I spoke about time allocations, like taking fragments of every single person, screenshots and everything that everybody's doing. And I'm building a dashboard that will show me, know, of person by person, activity by activity. It's basically taking everything we do, every conversation we have.

Craig (1:19:06)
Yes.

David Moses (1:19:22)
and it's drilling it down to major category, minor category entity, which is customer or property or whatever entity it's dealing with. And it's giving me some fits to build it, it's incredible what it's able to do. I think for me, I think it really is going to tell...

Craig (1:19:31)
Mm-hmm.

David Moses (1:19:44)
Where's our bottlenecks at? Where are people getting bogged down?

And I think that's a good, I think it's a good tool. It is kind of a significant shift for most small businesses to do. But if you look at small businesses versus large businesses, large businesses are hopeless. There's no way they're gonna be able to do something like this. It would take so much time and effort and money and small businesses, small to mid-sized businesses can do this where they can really drill down into the activities that people

are

performing and the conversations they're having really start to let bubble up to the surface where their bottlenecks are. I think for the first time in a long time, think small businesses have an advantage over large ones with their ability to shift into this new world and do so successfully.

Craig (1:20:31)
Yeah, to develop.

to develop tools that kind of help them level up with the big guys who probably aren't going to be able to catch up. I'll tell you what I'm working on that I'm really excited about and I'll have it ready soon. again, I'm going to be focusing solely on small to mid-sized lenders and hopefully extending them a line of credit for their lending businesses. That said, I thought an amazing marketing piece to them would be a, I've just pulled the last 50 loans

that your company has done. I know exactly what zip codes you're in. I know what addresses you leant on. And I'm going to be able to not only show you those, all of those on a map.

it by address and who the borrower was. But what are your competitors doing? What are your competitor lenders doing? Where are the houses that they're funding and who are their borrowers? And if I could give you something like that as a report each month for free, how valuable would that be in helping you grow your small and mid-sized business? It's literally generated just for you and your market. Right. So I actually have all of that data now and I want to put it in to the most amazing GUI.

That would be like a Google thing where, you want to see where these are all the houses that you lent on. These are all the houses that your competitors have lent on. And you can drill right down into them and go right down to the bar. We're level. And so. So that's going to be a Google that's going to be a elementics that's going to be linked in. I can get their LinkedIn content. I can get all of their their website.

David Moses (1:21:55)
Which APIs are you hitting to get that data?

but like the lenders and the

properties and all that data.

Craig (1:22:12)
Yeah, it's all elementics. Or for Casa, we're playing with both right now, and it's seamless. And so I think that's as a marketing piece to get the phone ringing for, holy shit, I just got this crazy email from you that was kind of a sneak peek into my market. What else do you have? Oh, well, I have this headache line of credit that we can offer you to hypothecate your loans. And you get this free report every single month if you're one of our lenders.

think it's great market Intel. It's really immersive.

David Moses (1:22:42)
And if you use the report correctly, you'll need us because you're going to have more loans than

you can fund.

Craig (1:22:49)
Yeah, and

so.

So adding on to that, I've developed sort of a really seamless onboarding process where frankly it should take them about 10 to 15 minutes to upload their docs and be on the approval process. And then if I can get the report that goes to the guy who says, yes, we'll lend to these folks, I think he could now instead of taking two days to approve, he can take 20 minutes to approve and we can start doing loans with these folks. So very exciting.

about that. I'll be showing that off hopefully on the next one if we have time and we hope you guys enjoyed this Fort AI podcast. Look us up the number 4kd.aifort.ai and we'll see you on the next one.