In this week's episode of Optimal Insights, Jim Glennon and CTO Seever Sulaiman discuss how Optimal Blue is deploying AI and machine learning across its platform, including a live demonstration of the newly unveiled Virtual Economist, a tool that combines large language model technology with four proprietary machine learning models to answer mortgage and economic questions in plain English.

Alex Hebner and James Cahill open with a market update covering oil prices, inflation risk, the ROAD to Housing Act, and Powell's final FOMC meeting. They share their expert opinions and insight into how these factors are shaping the industry and the broader economic landscape.

Key Points

1. The market update covers rising oil prices from the Iran conflict, inflation risk above 3%, a halved Q4 GDP reading, and the ROAD to Housing Act passing the Senate 89 to 10
2. Optimal Blue's Virtual Economist combines generative AI with four proprietary machine learning models to answer complex mortgage and economic questions in plain English
3. Optimal Blue has deployed more than 12 AI agents into its platform, with a focus on increasing development capacity without reducing headcount
4. Human oversight remains non-negotiable for AI-assisted code in financial engineering environments

Optimal Insights Team:

1. Jim Glennon, Senior Vice President of Hedging and Trading Operations
2. Alex Hebner, Hedge Account Manager
3. James Cahill, MSF/MSR Account Manager

Special Guest:

1. Seever Sulaiman, Chief Technology Officer, Optimal Blue

Production Team:

1. Executive Producer: Sara Holtz
2. Producer: Matt Gilhooly

Commentary included in the podcast shall not be construed as, nor is Optimal Insights providing, any legal, trading, hedging, or financial advice.

Welcome to Optimal Insights. I'm your host, Jim Glennon, Senior Vice President of Hedging and Trading Operations at Optimal Blue. Our clients and industry partners have long relied on Optimal Blue for trusted insights and commentary. And these podcasts are an evolution of our commitment to keeping the industry informed. Let's dive into today's episode. Welcome everybody. Thanks as always for being here. We have a wonderful show for you today. Whether you are an originator, a capital markets person, or just someone interested in

good mortgage industry information or just great market commentary, you're in the right place. We'll talk with James and Alex here in a minute. We'll do a little bit of a market update. And then after that, we had an interview with the brilliant Siever, our CTO and fellow innovator at Optimal Blue. We'll talk a little bit about AI, machine learning, and may even have a surprise or two in there that we'll be able to show you here in a few minutes. But first, let's go check in with James and Alex and see what's going on in the market.

Good morning, everyone. You got Alex and James here coming at you live on this ⁓ Monday, March 16th morning, St. Patrick's Day Eve. I James is ⁓ repping with a shamrock on the hat today. How you doing, James?

Yeah, pretty good. Happy St. Patrick's Day. know, between that and sinners kind of cleaning up at the Oscars last night, I was feeling, I feeling some representation.

Yeah, nice, saw, yeah, zero for nine for Marty Supreme I saw, so.

Yeah, you know, it was pretty solid movie. It's too bad he put his foot in his mouth.

It was, yeah, there

were some good ones this year. liked, liked Begonia personally. I actually have not seen centers, so I gotta carve out some time to go see that.

Solid, solid, you'll enjoy.

Yeah,

the Academy agrees with you.

Hahaha.

We'll jump right into it here with our econ update before cutting over to Jim and our CTO, Siever, who are discussing the landscape of AI currently. But just looking at the economic front, as a recap for last week, we got a PCE number. This was before all the geopolitical developments that have taken place over the last three weeks, RE the Iran War. PCE came in for February around expectations about 0.4 % month over month.

PCE is the Fed's preferred inflation metric. Sorry, I misspoke. This PCE number was for January. I would have thought it was for February, but it's for January because of the few-day government shutdown that we saw at the beginning of the year. We saw that gasoline energy prices, in addition to motor and vehicle parts, were the numbers that helped keep PCE this time around in check.

That's something to definitely keep an eye on because I think we can expect in the coming releases, not the February release, but definitely the March PC number that that energy cost will swing to the other side and act as an anchor to the upside for your inflation expectations. then I think the other big number that we saw last week was in the GDP revision for Q4. That number initially came out at 1.5%, but was halved down to 0.7%.

to Housing Act right here. They're just looking to expand the housing supply through a number of avenues. I think there's 40 plus different provisions attempting to get the housing supply into the hands of homeowners. Right now, the biggest point of contention is around institutional investors and how big of a share of the housing market they'll be allowed to hang on to. Right now, I think the

biggest concern is around for those that are considered large investors, those that hold 350 or more properties, they'll be required to sell those properties off over the course of within seven years of acquiring them. Is that correct, James?

Yeah, and I think it's good to finally get a little clarity on this one. Trump initially said he was gonna sign an executive order, try and get this through. People were raising questions like, well, no investment properties at all. Does that mean you can't have something that you're renting out in Airbnb where you have two houses, but you're not there? It's in Colorado or whatever, so you let people go skiing, rent it out at times.

⁓ It is no hey, this is for and not just large investors people who have a hundred plus. This is super large 350 houses plus so I think from my Perspective I think that this makes sense. It's going to be difficult to get it through both the house and the Senate This is definitely the sticking point But it's something that you know, you people go back and forth on well There aren't really that many of these properties owned

in the United States, but it is about 10 % of housing is wrapped up by private equity firms or owners. That is on average. can find places, Georgia, there's a couple of cities that have 25, 26, 27 % of the housing is owned. That's an enormous portion. is, you know, they're definitely making an impact on who actually can and can't be a buyer there. So even if they were just frozen out, that would make an them.

will definitely make an impact. But then, you know, are they gonna sell them at a little bit of a depressed price? Is it going to slow down if they're not allowed to buy? Well, will the housing builds in that area be at the same rates or would it be slowed? So there's a lot of second order effects here, but I like the initiative.

Agreed, agreed. Yeah, no, yeah. The forced to sell within seven years could, ⁓ you know, being forced to sell at that seven year mark, who knows where the housing market could be at that juncture. And it could definitely create, like you said, some second order effects and actually depress a housing market, especially in some of these areas where they have a pretty big hand on the scale in regards to that, that local housing market, like you said, in maybe the Atlanta suburbs or North Carolina, throughout the Sunbelt, I know is where private equity's snatched up the most homes.

So we'll continue to see how that bill continues to evolve. It's got to go to the House. The House has one that they're also currently debating. I think the short of it though is that Trump has signaled that he was likely to veto anything that doesn't include some sort of provision around institutional investors. So we'll just have to see what the heads in Congress, when they put their heads together, what they're able to put together.

Outside of that, like I mentioned, there's 40 plus provisions for increasing housing and getting more homes built, whether that's just making it easier to build, slashing some of that red tape. I've seen definitely some of the provisions are kind of encroaching more on historically what have been considered wetlands or river flood zones. then in addition to that, just ADUs, accessory dwelling units, kind of the grandmother's house out back in.

you know, parlance, just being able to have a rental unit on your property that you own. So we'll keep an eye on that. think that would have massive implications when it does pass. And it will pass in some form. This passed the Senate 89 to 10, I believe. So bipartisan approval across the board. addition to that, on the political circuit, ⁓ Friday, kind of snuck under the radar, Trump signed two executive orders on Friday.

targeting affordable home construction and the other was promoting access to mortgage credit. The affordable home construction quite a bit of overlap with what's currently being debated in Congress. And then promoting access to mortgage credit, just much more so they seem to be focused on getting smaller community banks into the lending space. And that was what that executive order was focused on. ⁓

first one there

kind of leans into what you were saying, right? I was reading through that too. And it was, you know, hey, the EPA needs to relook at their rules around wetlands and whether or not you can build on them. So can we relax a little bit of regulation red tape and actually get building started quicker? The second promoting access to mortgage credit, I think we're actually about to dip into it's a little bit more of a deregulation move, right? It's kind of rolling back some of the truth and lending act, trying to make it a little looser.

so that it's easier to get originations through.

Absolutely. Yep. That feeds perfectly into the next point that I want to talk about today, which is just bank deregulation. Michelle Bowman's kind of heading this up from the Federal Reserve Board. And this will be voted on this coming week with additional input still to come from focus groups and ⁓ groups that have interest in this space.

regulations for years now, several decades, really, really, since they were put into place in the wake of the financial crisis. And as I said, this will only really pull back these regulations to two levels that were in place pre-COVID.

mentioned a couple of times here, but I think what's continuing to drive the market and all the headlines that you're seeing day to day just revolves around the war between the US, Israel, and Iran and the straight-up Hormuz where 20 % of the world's oil and natural gas flows through.

We've seen the price of a barrel of WTI crude. That's the sweet crude that is pumped here in the United States that has increased by over 50 % since the war began from around $60 a barrel before the strikes began to north of $100 a barrel, settling around $95 a barrel at time of recording today. Seen a little bit of a give back there in the energy markets.

You know, there doesn't seem to really be an end in sight to the conflict and, know, the main point of contention here being at least for the markets around the straight forward moves and getting energy flowing again. I think that the main concern for you, you know, as a mortgage professional is just keeping an eye on what this is going to do to push inflation expectations up. You know, we've seen before the strikes, we were expecting two rate cuts from the FOMC this year.

cost per barrel, adding about five basis points to the core CPI reading. So where we currently are, we'd probably add about 15 to 20 basis points to the CPI reading, which would push us to right around 3%, if not just above.

it's a little relational into tariffs that this would be more of a one-time shock. We would see the hit if it actually goes up $10 a barrel somewhere in the 90s, maybe even $100 a barrel for a long time here. That's the five basis points that you're seeing push us into 3 % inflation. But at some point,

this would hopefully all be resolved and you would expect pricing to come back down. there is, you know, the big threat is if things continue to escalate, if a lot of that, not just the straight stays closed, but a lot of the infrastructure to get the oil out of the ground or transport at all is destroyed, that would be a longer term shock. It would be much more difficult to actually get everything moving again. And so that's really kind of the fear that's been stoked.

Absolutely, absolutely. Solving the conflict is just step one to getting global energy markets back into balance. The next step is then for all these refineries and the pumping stations to be turned back on, which has weeks of lead time. And if this conflict continues to drag out, there could be second order effects to where different countries source their inputs for oil, which...

when a country is building out their energy infrastructure, they're targeting a certain grade of crude oil. In the US, it's sulfur light, the oil that's coming out of the ground in the Middle East typically has a higher sulfur concentration, which all the refineries for these countries that buy this are built around. Yes, yeah.

You need to have specific equipment to work with. You can't just,

what's coming out of the ground in Venezuela and coming out of the ground over in Saudi Arabia or even the United States are not the same oil. And so you need different material and different equipment. So you can't just shop it around as easily.

Exactly, exactly. ⁓ We're not quite to that point yet in regards to the energy shock, but that could be if this conflict were to become a years long conflict, we would see those kind of pain points arise. So just continue to keep an eye on geopolitics and see if it continues to escalate. It does seem to have tapered off at the moment. At least that's what the energy markets are saying as of this morning, but that is changing by the hour, by the minute.

Just looking forward towards this week, again, I think geopolitics is going to continue to capture the headlines, but we will get a PPI reading. That'll be our last inflation reading for the month. The goods and services have been swapping off kind of month to month as to which one is creating inflationary pressures and which ones are seemed to disinflationary. Right now it's expected to come in right around where it's been at 0.3 % month over month for the core, 2.9 % year over year.

Again, this will be a pre-Aran War reading as well. So I would expect it to fall near those expectations. Then finally, we have Powell's final FOMC meeting that will also be taking place this Wednesday. As we already touched upon, there's no expectation for a rate cut.

Yeah, and I think I like your note there because it's his last meeting, but Tillis is still blocking Warsh's nomination. So there's not a clear succession yet, technically speaking in prior years. This is not the first time in US history that there was not an approved member to take the place of the Fed chair.

Generally what happens is the sitting chair remains in charge. It's not a written rule anywhere, so there will probably be some administrative pressure back in the news to try and figure out what exactly we're going to do here. Who you know? Does Powell actually get to stay and sit or does it just fall to someone else? You know, is there a way to force wash through so I would expect to see some headlines in next couple of weeks coming back around to this because.

the Fed is going to be critically important, especially now with inflation potentially back on the table due to the oil shock.

Absolutely, absolutely. think yes, I think it's very pertinent to caveat that with an asterisk. It's his last official FOMC meeting. There's a chance that he continues to carry on his duties beyond his, I believe it's May 15th. is again, asterisk final day. Yes, we'll continue to keep an eye on that. I know last week a judge kind of rolled in in pal's favor in regards to the ongoing ⁓ criminal allegations in regards to the FOMC's. ⁓

Yeah.

official last day.

redoing of the Eccles building and adjacent auxiliary campuses. We'll continue to see again how the Trump administration's efforts to prosecute him continue to evolve over the next two months here.

Something to look forward to.

Some look forward to it.

Perfect. I know that was a lot of information. We were blessed by the news this week in regards to content. with all that being said, we will hand it over here to Jim and Seever who are walking through some AI and how we're using it here at Osmo Blue and in the mortgage industry. Thanks, James.

Thank you.

All right, welcome Seever. Seever is Optimal Blues CTO and long time friend and coworker of mine. Welcome Seever thanks for being here.

Thank you, Jim. Good to be here.

We wanted to have Seever on today. It's been a while. We talk tons about AI on this podcast, right? You can't go anywhere in the industry, even in the world without hearing or talking about AI. We've had some of our own experts on the podcast. We've had third parties come on to talk. And of course you hear Alex, James, I, Kevin go off on tangents about AI almost every week. But I wanted to bring you back in Seever just to kind of level set us. feels like...

Yeah, of course.

We're in the very early stages of AI implementation across the world, certainly in the US, certainly in the mortgage industry, but it feels like we're seeing some benefits from it at this point. Where do you feel like we're seeing the biggest and most immediate benefits of AI in the industry, but also inside of Optimal Blue?

Yeah, for sure, Jim. And it's actually, mentioned that you talk about AI all the time. One of my team members during the Optimal Blue Summit last week, they counted the number of times AI was mentioned. and Gen. AI during all the conversations and the speeches, not just from Optimal Blue employees, but also from all of our guest speakers, the panels. It was something like 624. And it was just like the most mentioned word.

Mm-hmm.

more than any other sort of unique individual world, word in the entire summit. was pretty interesting. But I think when it comes to, um, the utilization of AI, I do want to differentiate between, and we've talked about this last time, right? Between GenAI and AI. know we will touch on this a little bit more, but specific to generative AI, which as we know is used for completions, for language processing, et cetera.

There's really different layers where it's being utilized. Obviously in productivity, as you know, where we are and many in the industry or some in the industry are using it for productivity, whether that is just employee efficiency, if you're using things like Office 365 Copilot or what is more complex in sort of coding tools, coding, testing.

I mean, software coding, right, development. So that's really one layer where you see some of the organizations talk about how they're generative AI in code assistant. The other one is really in the products, right, in the workflow. How do you help your customers, the mortgage lenders, the partners, the brokers, in their workflow and in their processes by...

deploying what we're calling AI assistants or AI agents. And these are GenAI agents that go in. And as you know, we have so far more than 12 of those, like the profitability assistant, the position assistant, obviously the Ask OB agent, the originator assistant. All those are features or products that we have delivered in the market that help customers with their individual

Mm-hmm.

Mm-hmm.

processes and workflows. Originator Assistant, for example, helps loan officers find better rates for their customers or the borrower. So to answer your question in really one sentence, it's kind of being used in multiple layers. most that you hear about is encoding in software development. And I think we are seeing more and more companies like us implement Genactive AI as agents to help.

our customers, the lenders, the brokers, do their work more efficiently.

Right. So on one hand, you've got, you know, we hear it from some of the software companies already, even though there might be a little bit of noise around whether they're laying people off to right size or to try to utilize AI just to become more productive, which is what I believe we've done at OB, which is to just be able to move faster, right? By having these AI tools help us code, but we're also deploying then tools for our customers, like you said, which is whether it's...

That's exactly right.

And you know, as far as, mean, what a lot of people talk about sort of layoffs and unemployment rates related to gen AI, to sort of what we're doing, at least in the industry or for Alkali Blue, we have started utilizing generative AI in our software development process. And this was done more than a year and a half ago, or maybe we started You probably hear like every week something new comes up, right? There's new releases.

There's a race. There's definitely competition between OpenAI, Anthropic, and the prop cloud. Now we have Croc. So these are all trying to improve their LMs for specific use cases, not just for coding. So there is a lot of new advancement that comes out every week. But for us, where we have implemented it and we continue to implement it, is how do we increase our capacity? How do we increase our devolving capacity?

so we can deliver more for our customer, we deliver them faster. you know, Jim, we have a product roadmap that goes for at least a year, year and a half. That is what is already laid out and the vision and the ideas go even beyond that. So my thought is always, how can we deliver not only on that, we're working on our Q1 and Q2 roadmap today, but can I get our development organization in a position where we can start thinking about...

working on Q3 items. And we can do that by increasing capacity. And as you know, we can't just go hire 20 people and have them be effective day one. But if we're able to utilize AI to help us increase our capacity, even by 10%, by 5%, that really helps us with our customers. So that's really what we're focusing on is how do we help increase capacity and quality at the same time? Because you can use Gen.AI not only in coding.

Mm-hmm.

but also obviously in the testing process. It can help you test better, find more issues faster before anybody else finds it. So that's really our focus from a software development lifecycle perspective is increasing capacity while increasing maintaining quality, but also increasing quality.

Mm-hmm.

Right. We've called it the capacity explosion, right? There's a few components to it, but it's ⁓ largely driven by AI and just our ability to get features done quicker, which to me has always been kind of the more optimistic side of what happens with AI. It's not to be able to do the same amount of work with less people. It's to be able to do tons more work with the same amount of people, which is how we're employing it today.

I was ready.

That's right.

That's exactly right. That's going to need capacity with the same amount of people. So it doesn't replace the job because you still need that, you hear about human in the loop, especially when we have senior architects or senior developers. I don't want any code going to production without my developers looking at it. Because what we do in financial engineering is really critical for our customers. we're not.

Any near the point where we're going to trust AI to build code for us, do code reviews, test the code, test the product, and deploy it to production. I don't see that happening for another two years, especially in our industry, or at least our optimal loop, because accuracy, as you know, is very important to us. That's really important to our customers. So we deal with product appraising or

MBS pricing, trading, we can't have any errors. So that accuracy is critical for us. So we're not going to have written by AI, be deployed by AI with nobody else.

Right. And presumably that's going on throughout our industry and throughout the world, just in the way that it's being deployed, the way AI is being deployed. And it makes sense. You can't have the same group of people develop code, deploy the code and test the code, right? You always need a second or third set of eyes. In this case, it would be the set of eyes of a human before code is actually deployed to production because there can be mistakes. mean, I see AI hallucinate more than it gets the answer right in some applications.

Absolutely.

If we're inventing new technology, the last thing we want to do is let it automate itself and potentially cause problems.

100%. Yeah, 100%. I we have senior architects using, for example, Claude Sonnet, right? And they use it for one problem or one project. It works perfect. The next one, doesn't, right? And they know because they're looking at the code. And that's where I trust my development organization a lot more than I trust AI. But we are going to, we are using GenAI to assist us get a little faster, right? There's some...

Mm-hmm.

work that you can actually have like unit test ratings, stuff that you can have AI built and write for you, but it doesn't really replace my organization.

It does some of the heavy lifting, the manufacturing of code, if you will, but it's still supervised. that's, you know, something that leads us to something else I wanted to touch on, which is related to AI. It's a specific application of it, I guess, is how I think about it, which is machine learning. And in our industry, there's a lot of different areas that you could envision deploying machine learning. There's several areas where we have already done that. So to me, machine learning is...

It can relate to AI, I think, because it can learn. It learns from new information, but it generally is for applications where you're looking to chew through tons and tons of data to look for, whether it's trends or anomalies, mistakes, or even the ability to project what may happen in the future, whether it's you're trying to project volume for your organization as a mortgage lender or interest rates in order to decide what interest rate policy or even actual

market conditions might look like. Is that, do I have it right there? First of all, like, that generally what machine learning is used for and how we would think about it?

Yeah,

absolutely. I machine learning is artificial intelligence. I mean, that's really one of the core of artificial intelligence. And all of these LLMs that you hear about, which is the foundation of generative AI, the LLMs are machine learning models. They're actually built, trained, using machine learning techniques and algorithms to build these LLMs, which are static.

models deployed. So you hear about GPT 4.1, GPT 5. What that is, is really it's a trained model using machine learning neural networks based on, I think GPT 5 has 1.2 trillion parameters. Some of the smaller ones has like, know, 16 billion parameters and that's considered small, know, SLMs. So these trillions of parameters go into this neural network machine learning algorithm to build

a model, right? And that model is your language language model or the LLM, which is really a static model that is used and deployed. And that's what we all use. know, Sonnet is one of them, obviously GPT, et cetera. But machine learning itself is exactly what you said, right? It could be predictive. It could be classification where you feed it a lot of data, a of data, and you build a model that becomes sort of like a

like an engine and that model becomes a process where you can feed it data and gives you data. You input with some information and it gives you an output. So for example, if you build a predictive model for let's say bulk bit pricing model. So you can feed it a ton of data to train it based on historical information of how do you invest those beta loans and then that becomes your model. And then at runtime or...

Mm-hmm.

When you want to use it, you say, okay, here's a scenario that I have produced what the expected value would be or the expected bid price would be. That's an example of predictive modeling or regression analysis. And there's different obviously types of regression analysis, linear or nonlinear. But there's also models, non-predictive models, you some classification where you feed in a ton of data and that's really what comes with LLMs.

LLM is both predictive and classifications because you give it all this data and it recognizes and it tokenizes all that information based on the, like I said, trillion of parameters that go into the LLM. But also then we build a predictive model that would predict when you're going to say next or what comes out of next. So when you are giving it your text to summarize for you, for example, it has the ability to do that based on

all of the information that was fed to the model. been utilizing machine learning, but recently sort of formally established our own machine learning organization, as you know, to work on how we can take advantage and utilize the data that we have within Optimal Blue. Because as you know, we have a lot of information, we have a lot of data we can use to build proprietary predictive models that help our customers.

Mm-hmm.

Fantastic. Yeah, it's super fascinating. you use the example of bulk bid prediction. That's something that we've actually done for years here. And Optimal Blue, in case folks aren't familiar, if they don't work with us on the hedging and trading side of the business, we've been able to act very accurately predict what bids look like for live bids on mortgage loans. you could see how with anything, with sports, like predicting sports winners, predicting the weather, like a lot of these models are being trained.

to again, suck down all this historical data, look at what may have changed, whether it's in the market or in the weather and manufacturing or whatever your area of business is, and predict future outcomes, which to me is just fascinating where normally it would take a human or a group of humans or actuaries to compile all this information, try to determine statistically significant trends, and then somehow get it right every time in terms of their math and their predictions.

That's right. That's

really the beauty of machine learning is it is a pillar of AI and it's proven because when you build a model, you always come out with a margin of error. know that there's multiple metrics that you use for machine learning model, like R-square or mean square error, but you have a number, right? You know that the accuracy of this is, say, 98 % and you have a margin of error of 2%. But the more data you feed it,

And the more the data, meaning that you have different data sets or differences in the data, the more accurate and more predictive your model is going to be. Because you're not going to overfit it, you're not going to feed it the same information. And it's always going to say, yes, that is correct because it's same data. the thing about machine learning is you can really rely on it. And obviously, in financial engineering models, there's

People rely on it. Netflix is an example where it predicts what you want to watch next. And most of the time, it is accurate. It is not always accurate because there is a margin of error. is a 2%, 3%, 4 % error rate. But it is published with these MLMs, sorry, with these machine learning models. Say, OK, the confidence score is 95 % or 98%. So you know what you're expecting out of

Pretty good.

Mm-hmm.

All right. Less accurate when there's a human component to it, like my own preferences of what to watch on Netflix, but probably a higher percentage for things like reading documents or maybe predicting volume. So yeah, I want to tease a little something here, something that we unveiled at the summit. So many people got to see it. It's not yet in the hands of clients, but we have a full working model of it. It is a machine learning AI assistant

That's right.

We call it the virtual economist and it's exactly that. I think it's very well named. I want to see if we could pull it up on the screen here and ask it some questions. So this is an AI, presumably large language model AI feature that also implements and uses machine learning to make all sorts of predictions based on data that it can go out and find instantaneously. We talk a lot on this podcast about...

what's going on in the economy and how it might affect rates going forward or how it has affected rates this week. We've talked a lot about like the conflict in Iran, for instance, and how that might change what's happening right now in the economy. frankly, Alex and James and others do tons of research online, listening to podcasts, reading articles, sometimes using things like Copilot to go out and find good information. But this AI assistant basically goes out and does all of that for you. Just if you ask it a simple...

question in plain English, which I just thought is just one of these kind of perfect applications early on for AI and machine learnings.

Right.

I this is where we marry AI and generative AI, because it has at the top layer, it has the generative AI where we do use the LLM. We didn't build the LLM. These are obviously the GPT models that we use in this case. But it does use that for natural language processing, for speech recognition. But also, you can ask, and you will pull it up, where you have an avatar. Those avatars are also.

Generative AI built and you speak with the avatar and it's able to process the information using Generative AI, right, using the LLMs in the backend. But it's also able to respond to questions based on that. But the beauty is the foundation of its knowledge is machine learning models that we built. There are four models currently that we built in addition to the OBFMI data that we have.

also fit to it. So those four models are proprietary models that we built for interest rate prediction as well as rate lock prediction based on optimal blues data. And that's where we kind of combine Gen.AI and machine learning, which is also AI, into one application that is this virtual economy.

Right.

Right. So it can use data from everywhere, from out on the web, where we may ask it for current events, then it can internally pull from our optimal blue data.

So down the road, you would be able to ask questions not only on publicly available data and trends, but also tie that back to your own pipeline within your own business.

That's right.

Cool. Okay. let's go ask the virtual economist a couple of questions, shall we?

Let's start with something a little more involved. How we ask it how projections could change if we see some changes in Fed policy. So we'll ask what are mortgage volume projections if the Fed cuts six times in 2026?

Cool. Very interesting. All right. So that's obviously taking from some of our internal data and making projections based on also some market data that's available to the economist out there. How about we go microphone here? We'll ask it maybe what...

What are some possible effects on the economy stemming from the war in Iran?

That is just wild. It would have taken quite a bit of research to come to some of those bullet points.

Alright.

Let's ask the economist one more question here.

find out what its thoughts are on the incoming Fed share.

Very cool.

What are your projections for the OB-MMI for the rest of 2026?

All right, very insightful. Well, wonderful.

Thanks again, Virtual Economist. Thanks again, Siever. Talk again soon.

All right, let's wrap this thing up. Thank you so much, Alex and James for the market update. And Siever, wonderful interview. Thanks again for doing that. We'll talk to you again here real soon. And that's it for today. Join us next week for another episode of Optimal Insights, where we'll continue to provide you with the latest market analysis and insights to help you stay ahead. Check out our full videos on YouTube. You can also find each episode on all major podcast platforms. Thanks again for tuning in to Optimal Insights.