The technology conversation of GenAI has moved so fast since OpenAI and Chat GPT burst into the public eye. But are businesses just buying into the hype of AI, and does it really benefit their goals and needs? For Shashank Shekhar, Founder and CEO at InstaMortgage, he is taking AI's capabilities and augmenting them for mortgage and lending companies to solve challenges specific to the industry to improve efficiencies and productivity. What started as a curiosity has turned into creating a tangible way to positively impact business operations for loan officers, processors, underwriters, and other key stakeholders in the lending process. Learn more about how InstaMortgage builds on large language models and trains them to become more relevant to the mortgage industry in the latest Expert Insights podcast episode with Total Expert Founder and CEO Joe Welu.
Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.
Transcript
From the show. Tap a time to play from there.
SPEAKER_00Hello everyone, I'm Joe Well Yu. Thanks so much for tuning in to Expert Insights, where we talk with industry leaders across modern financial services to discuss leadership and innovation. Let's get started. Hey everybody, welcome to the podcast. I'm super excited today to be joined by my good friend and industry leader uh as well as AI expert, uh Shawshank Shakur. How are you doing, brother? I'm doing great, Joe. How are you? Good, man. Good to good to be with you and excited for the conversation.
SPEAKER_00Uh, just for the audience, uh you're founder and CEO of your own mortgage firm, InstaMortgage, and you've also been incubating some really impressive AI technology inside your organization. Maybe just give us a high-level uh education on what that is and what that looks like, and then I've got some questions for you. I I think the audience is going to be interested in.
SPEAKER_01Yeah, I mean, when when really OpenAI came up with Chat GPT uh what early last year, actually, that's when it started intriguing me the entire concept of AI and how it could potentially change a lot of things within within the mortgage space. Uh, but the only way to test it was, as you mentioned, to incubate it with with my own mortgage company, InstaMortgage year. And and that's how I kind of it was it was more out of curiosity than anything else, is saying, okay, these capabilities of Chat GPT, for example, how does that translate into a mortgage space?
SPEAKER_01And how does that change or improve, hopefully, the efficiency and productivity here at our company? So that got me started, that got me curious about the entire scope of AI. And that's practically so what's been about 12, 14 months that we have been working on some of those things.
SPEAKER_00Yeah, that's great. So we're gonna be going deep on this for just for the audience's sake. Uh you and I are gonna have a we have a session, uh, fireside chat session at our conference, Accelerate coming up in a couple of weeks. Really excited about that. So we're gonna talk a little bit on the surface about a few of these things, but uh I know we're gonna go super deep on some of these topics and your most importantly, your key learnings as it relates to how to how to set up AI to be successful and impactful in the company, not just do AI projects for the sake of hype, right?
SPEAKER_00Uh I think you agree with me that we've seen things moving incredibly fast. And there's a point of view as it relates to Gen AI, and one of the phrases that I've heard oftentimes used from CEOs as I'm talking to them about the power of this technology is you've got hype and you've got actual real impact. And what I love about how you have focused these use cases and the adoption of the technology is you're doing it from a point of view of somebody that has real business problems to solve in the organization, and you're doing it based on impact, not just putting a AI phrase on it.
SPEAKER_00Um AI covers a lot of ground. And how how did you decide on the specific types of models to use? Like walk us through kind of how you got started and your point of view.
SPEAKER_01Yeah, as as I mentioned, Joe, a lot of it was started with curiosity, then kind of went into experimentation and exploration as to uh because when a technology is new, you have to be open to be doing all of that stuff. Is because unless you try it, unless you test it, unless you experiment with it, you really are not able to get to the point where you're able to figure out what is important, what is impactful. The the word that he used is very important, is because not just within the industry, but across other industries too, you are seeing CEOs who are going from uh say absolute denial, which is AI is not going to do anything to my business at all, to even panic, saying that whatever I do, how does it matter?
SPEAKER_01Uh AI is going to going to going to replace me anyway. We are looking at AI at somewhere in the middle where it can positively impact uh our business. And when I say our business, I mean everyone in the entire food chain, all our stakeholders. We are talking loan officers, technical processors, underwriters, everyone in the process.
SPEAKER_00You're really talking about adding productivity and enhancing their ability to get the jobs that need to be done more effectively, faster, um, more accurately in some cases. There's what you're talking about, right?
SPEAKER_01That's that's correct, too. And and what we're looking at is really Gen AI is what you mentioned before. That's I mean, AI has been here since, I don't know, 18, 70s. Um, I actually know someone who did PhD in AI in the early 80s from Cornell. So I did not even know AI existed back then. But AI, in a sense, machine learning has been there forever. Really, what has changed in the last say 18 months or something is that evolution of Gen AI. That OpenAI, of course, started with that. Now you have models like Gemini from Google and Claude from Enthropic and Lama from Eda.
SPEAKER_01So there are multiple different companies. And what it has done for people like us is we could never have had resources to build a large language model that these companies have provided us. Uh, what we are able to do is now build on top of it, use some of the models that exist, and then train it specifically for our industry, in this case, of course, of course, mortgage to build it. And that's a big game changer, is because now companies like us and people like us can actually use AI because of what OpenAI has done, because of what Google has done.
SPEAKER_00And it's ultimately what you're describing is you don't you don't have to spend the millions, tens of millions, hundreds of millions to build and train your own models. You can now start with some foundational open source models and then train them for your specific use cases in the industry. And that that's really been your approach. Like you guys are leveraging the the large language models. You're not, just to be clear, you're not developing, and if you are correct me, you're not developing your own model per se.
SPEAKER_00You're simply optimizing those existing models with the data and the use cases and the information that's relevant, right?
SPEAKER_01That's correct. So there are there are really three kinds of companies in this space. One who have created the large language model. So we I mentioned four or five leaders in that space. Then there are companies who are just using the large language model the way it exists. And that's the that's the low end of using, say, something like Chat GPD. And you would see that in chatbots, you would see that across multiple different companies. Uh, but really where I see the right fit and where where InstaMortgage and Insta AI comes in is using the large language model, but to train that on um on, say, our expertise, in this case, in this case, the mortgage.
SPEAKER_01So you're training the model uh to be more relevant uh for the mortgage industry. And that requires a huge amount of training, by the way. We are talking um thousands and thousands of pages of domain expertise that you need to train and retrain and iterate and figure out where it's hallucinating and then keep building that process.
SPEAKER_00And when you say training, really your ability to get uh really high impact outcomes out of the case of the technology is going to be proportional to your ability to train it with very relevant specific data. That's that's what you're describing, correct?
SPEAKER_01That is correct, Joe. Is because the way we need to understand how large large language models are built. Uh we we use the LLM board quite lightly. What we do, what we need to understand is that large to build a large language model, you have to train it on everything that exists out there. It's like the the the knowledge of the entire humanity, so to say, is in OpenAI or Chat GPT here. But the problem with that is that it's almost like a jack of all traits in Master of None kind of a thing, is that yes, it's really good at few things, but it does not have the same deep domain expertise, say, on mortgages or real estate or insurance or financial services.
SPEAKER_01Uh that's where Morgan Stanley, for example, built that generative AI tool in September last year. They partnered up with OpenAI to build uh what they call financial services assistant. And so that's where the that's where the huge opportunity comes in is that you take a model like that that gives you some of the technological advantages, but then you build your uh say subject matter expertise or domain expertise on top of it. So our one of the inspirations for us, of course, came from what Morgan Stanley did. Uh, I mean, I personally believe that we have built something even better than that, but that comes from building the domain expertise on top of large language models.
SPEAKER_00Yeah. It you know, as is uh someone who's done a lot of research in the space and and obviously um internally at Total Expert, we've got lots of different things happening in the AI front, but there's been so many key learnings over the last 18 months or so. And I I think the essence of what you're describing to me is so much of the value creation from this technology wave, this in a really revolution in a way, is going to come from inside the walls of the enterprise and their ability to harness these things in a higher level of specificity in ways that are mm super relevant to that domain.
SPEAKER_00Right. I mean, and that's your vision as well and your key learnings from the way it sounds.
SPEAKER_01Yeah, and that's what I mean. Of course, some of these companies had early advantage because they partnered up with OpenAI. But if you look at Duolingo, uh, if you look at um, say, uh Khan Academy, if you look at Morgan Stanley, that's what they did. They took the large language model that that OpenAI built, but then they put in their expertise on top of that uh to create meaningful impact. And that's what we are looking at doing, uh, not just from generative AI perspective, but we are looking at say other, some of the other artificial intelligence um technology, like say computer vision technology, which has become significantly better in the last 12, 18 months in terms of how better it can read, say, multiple documents that we create in the mortgage and real estate space.
SPEAKER_00So high-level advice. We're gonna, as I mentioned, go deep into some of your key learnings and uh the uh problems you're solving, the value you're generating internally, which I gotta say is uh the most impressive as a visit with executives um was on uh over the last couple of days with several different firms who have different AI initiatives internally. And um, it's super impressive that you've been able to get value created so quickly that's super tangible. And if if you think about how fast things are moving and the pace of the evolution, how do you suggest companies approach in implementation of AI capability, gen AI capabilities in general, considering that everything that was true yesterday is going to be different a month from now?
SPEAKER_01Yeah, and the good thing is that it's still true, it's just probably better true. I mean, if the last thing that that that I would want anyone to do if they're serious about this, is just to sit there and just see the the evolve, the evolution of AA happening, right? The last thing you want is this is say we're built, we are creating, say, uh a skyscraper here, and we are on ground zero, maybe first first or second floor now. But you're waiting for this to build, and they're like, okay, we are we are on the tenth floor now, let's get into this space.
SPEAKER_01Uh and that's why we brought in the entire experimentation approach, more a startup culture, so to say, to this entire approach is like, let's test it, let's work on it, let's see where it goes. And the reason for that is if you truly believe that this is probably one of the biggest disruptors that has ever happened in the last century, then you need to be in it very early. The earlier you are in the process, the more you understand it, the more you have had the time to iterate it at a much low cost, uh, before you are say, before you lose the competitive advantage of having AI by your side.
SPEAKER_01So that's how if I were an executive and an I am one, that's how I looked at it is that I don't want to be looking at this in 2026 and thinking, why did we miss this massive opportunity that we had?
SPEAKER_00It's so interesting you're saying that because I do genuinely see that mentality showing up. There's really two rules of thought, which is let's go now, let's get out of the gates, let's find the fastest path to value creation and experiment and ideate on those things. And then there's the other uh sort of group of leaders that are saying, I want to see how this shakes out, I want to better understand data and security and what's happening with my data. And do you think those people are just not as educated on really how fast things are moving and the power of this technology, or do you think it's something different?
SPEAKER_01No, education plays a huge role. I mean, I've been I've been at executive roundtables, I've been I've been uh talking about this to other people, and just the uh I think there's ignorance to to a certain extent of what they think AI uh AI can do in a bad way, and it's just not the clear understanding. I mean, we there are ways, I mean, just for the audience to know, is that there are ways to build stuff on AI with complete privacy and complete protection of data that that exists today, let alone tomorrow, uh, the the ways you can do it.
SPEAKER_01So also we are already thinking, um, but what if AI makes a bad decision, say on a loan? Uh, first of all, breaking news humans do it already. So it's not as if that that we are not doing.
SPEAKER_00You're saying salespeople or loan officers or advisors might say or send things that aren't actually accurate anyway.
SPEAKER_01Yeah, our processors are underwriters. So I mean this huge really, there's this paranoia about the fact that oh, AI will make a mistake. Uh, first of all, we are not using AI for decision making. We are not going there yet. We will eventually get there, by the way. Uh, but um, even if it did uh make a mistake here and there, I mean, that's where human in the loop is. We are not talking about eliminating workforce because of AI. We are talking about improving their efficiency, improving their productivity. So, yeah, you're right. A lot of this is is about probably listening to the right people on what AI is capable of.
SPEAKER_00I I think you you summarized it pretty well. And if I think through some of the conversations we've had over the last several months, and you guys have had just this point of view and perspective and and really ultimately a vision on first of all, where do I see this going generally in terms of the technology? And you grounded really quickly on this is gonna be big, it's gonna change everything. And then you you formed a thesis on the way to approach it for your business.
SPEAKER_00And I feel like where executives are getting a little bit lost in the velocity of things evolving is they're overanalyzing some of this almost to a fault versus having really tightly controlled teams, really cut tightly controlled ways of um training the models with your data, just putting small, really lean teams around uh projects that are experiments, essentially, and not saying, hey, let's go implement this end-to-end across the organization.
SPEAKER_00Um give us a framework or or uh a current point of view on how you have done that. Are you still really in the experimentation phase or now you in the I'm I'm rolling, I've seen enough value, I'm rolling it out really rapidly.
SPEAKER_01We are all in um on AI, for sure. Uh and that is uh and that is not to say it's because it's a buzzword or some everybody's talking about it. It's because um we are thankfully about 12 months into it, 12, 14 months into it. So um excuse me. Are we into experimentation phase? Yes, at all points in time. We will continue to experiment with something new. Yeah uh I don't know for how many years.
SPEAKER_00But we have sort of if you're a true innovator, you never really leave. Exactly. You do that, yes. Yeah, right?
SPEAKER_01Just like, yeah, you see, I mean, product releases from Total Total Expert happening all the time. It's because you are always experimenting. But that does not mean that you already do not have things that you that's already in production, that's already making an impact. So it's a simultaneous process with with everything. Um, but I think we are in a phase where we already have products which are live. We already have products that's making an impact while we are still experimenting with other things.
SPEAKER_00I don't want to go too deep into it, but I do want to give the audience a little flavor for some of the use cases high-level. Uh we talk a lot in generalities around the space, and I know people love to hear specificity, right? Anything you're willing to share, maybe just give us a little bit of a preview of what we're gonna talk about at our conference.
SPEAKER_01Yeah, and the way I see uh AI impacting is, as you said, wing-to-wing, really, is because the way we are looking at AI is three stages of the loan production. Uh, we are looking at uh pre-processing, which is where we are looking at loan officer efficiency and productivity, how we are improving that, how are we helping them market themselves better, how we are helping them develop and improve their skills better. That's that's what the core of Insta AI is. That's a Gen AI platform. Then we are looking at loan production, which is probably the biggest challenge in the industry uh right now in terms of how high the production cost is.
SPEAKER_01And that's where we are looking at computer-vision technology and large language model to solve some of those efficiency problems. And then we're looking at postclosing is that what kind of predictive analytics you can build in uh using machine learning to figure out uh what do you need to do client, uh to do client retention. So we and that's where when I said while we have something live.
SPEAKER_00The last part you just described, that sounds more like that's traditional AI, machine learning, the way we thought about it, not generative AI. Maybe just spend a little bit uh for those of the audience that are listening that don't necessarily know the difference between gen AI, regular AI, machine learning. Maybe just give a little bit of a description on what you're referring to there.
SPEAKER_01Yeah, I would love to. And and that's why I said the AI has been here since probably the 80s or 70s. It's it's just that it has become a buzzword uh because of Chat GPD primarily, is is that a concept like machine learning where you are teaching the machine or you're feeding the machine large amount of data, past data, for it to be able to model future or what is called predictive behavior, that's machine learning, that's artificial intelligence. Uh it's like if you had if you had, say, I don't know, last five, six, seven years of if you're a mortgage servicer, that kind of data, and every time they defaulted, you of course recorded that.
SPEAKER_01We take all of that data, put it into a machine learning model. Now it will be able to predict which are the kind of bars will potentially run off if the rates were to say be half a percent lower, uh, or if if, say, there was a recession. All of those things is is uh what you call traditional artificial intelligence, uh, something that we have done. Even something like OCR, uh, which a lot of us have used in the past, just image reading or image recognition, is part of computer vision technology, which is still a traditional artificial intelligence. It's just that even machine learning and computer vision technology, which is technically traditional AI, has evolved substantially in the last say, say, 16, 18 months.
SPEAKER_01Uh, and some of the lines have been blurred. Actually, uh, what say GPT-4.0 does is there's a lot of advanced computer vision technology there, uh, which is we think about it traditional AI, uh, but but that's what they're doing. Generative AI, on the other hand, it's something that we have never seen before. It's the ability of artificial intelligence to create something out of practically nothing. Uh, probably something that we have never seen. Create content.
SPEAKER_00That was literally literally, right? It's assembling something from, you know, we're looking at some of the experiments we're doing on the content side of things. It's literally creating new assets and content that didn't exist before. Yeah. But it's doing it from the lens of your brand, your voice, the things that are relevant to you because of the inputs that have been used to train it, right?
SPEAKER_01That's the game changer, really, is that? I mean, we have seen, of course, on the fun side, we have seen Jedi creating amazing videos and amazing music. And they're the written lyrics that some of them are saying is better than some of the some of the There's some really cool magic tricks out there, isn't there? Yeah, exactly.
SPEAKER_00But yeah, that's that's that's what we'll be discussing at at Accelerate, really, some of these things that we have built and how they're improving uh the Yeah, I'm I'm I'm super excited to to go in uh to go into those actual use cases and give the audience some advice there. Uh for for those that are in the executive, the C-suite chair, or running companies like you run companies, what would their expectation should their expectations be in terms of do they need to think about recruiting a different cohort of talent into the IT side of the organization?
SPEAKER_00Or would you say it's more about selecting the right partners or a combination of the two?
SPEAKER_01A combination. I mean, it really depends on uh what your need looks like and and what uh what kind of budget and and and manpower do you have uh at your disposal? And in our case, really, uh we built it on shoestring, but it also was a lot of curiosity, uh which a lot of us lack, unfortunately, in the industry. But if you come up with that, uh you have a very clear sense. I think you mentioned about this earlier in the conversation, a very clear sense of what you want to get out of it, not just hype, but Real impact is what you're going after, then yeah, potentially you can you can build it yourself.
SPEAKER_01Uh it does require a lot of knowledge, uh a lot of education in that field.
SPEAKER_00What I found super fascinating about uh Insta specifically is how much you were able to do. And I know I won't share the numbers or I don't know have the exact numbers, but generally I know the how lean you uh were in terms of funding this project and somewhat intentionally, I think, in terms of saying, hey, before I go and throw huge cash amounts at this, I want to iterate my way to finding value really quickly. And so even from my point of view, and I live in the technology space every single day and every hour of every day, really, and I was blown away at what you were able to do and in such a small budget and have actually taken that to some of our internal teams, and I'm like, this is inspiring.
SPEAKER_00Okay, like you can really create impact with this technology, and it doesn't, you don't need an entire scrum team everywhere on every side of the product to do it. You can do it really quickly with some rapid prototyping and some rapid iteration with a very lean, lean team.
SPEAKER_01Yeah, I think Sam Altman said it uh said it very wisely, of course, is that you can have either three months of iteration cycle or you can have three hours of iteration cycle.
SPEAKER_00That alone saves a huge amount of cost. Yeah, it is really about the pace of iteration. And I think uh, you know, we we have both been in financial services space for a long time. And I think the thing that you and I both agree on is a lot of times people have just become their own worst enemy in terms of they need to get out of their own way sometimes and think you're you're slow-moving, massive legacy industry, legacy organization.
SPEAKER_00And you sort of just have to sh reshape your mindset in terms of the power of what's possible. And uh, I think you exemplify that exceptionally well.
SPEAKER_01Yeah, it's a story that we tell ourselves. It's like yeah, I'm not if if if I do not want to use AI, these are the stories that I'll tell myself.
SPEAKER_00Is that like we don't control it?
SPEAKER_01I don't know what it does, I don't know what will happen to security.
SPEAKER_00You can always find an obstacle that will prevent you from innovating. No, and uh there's no shortage of that as you go around you know the room in most companies, and uh it's the ones that challenge that and say, we're just gonna make things happen, do it smart and intelligently, but uh and make sure we're protected, but ultimately move and take action quickly and in and get things done so that we have a place to iterate from, versus, like you said, waiting a couple of years until the building's built and the plane is gone and you're left in the dust.
SPEAKER_00So Shah Shank, uh really love the conversation. Uh, you and I could set and and talk about the possibilities and how this is gonna change our our businesses and really ultimately change the world uh for a long time. But I think rather than do that, we're gonna save some of the deep dive uh for a couple of weeks from now at our conference. And for those of you that are not able to attend our conference or or um uh for whatever reason are not attending our conference, we are going to record that session as well and look forward to uh packaging that out in a podcast format uh following that.
SPEAKER_00So uh Shaw Shank, um congrats on all of the success that you guys have had this year on the innovation you're doing around Gen AI and AI. And uh really looking forward to seeing you in a couple of weeks, my friend.
SPEAKER_01Yeah, and loving the concept that that we have had here and looking forward to our conversation.
SPEAKER_00Thanks so much for joining us today. Be sure to subscribe wherever you get your podcast.
American Pacific Mortgage is on track for $15B in production volume this year. What does it take to lead an organization of this size through the biggest technology shift in history in the most challenging market in a...
Ever wondered how one loan officer can build a billion-dollar business? In this episode of the 10x Lender Series, Total Expert Founder & CEO, Joe Welu, sits down with Shant Banosian of Rate to pull back the curtain on...
At the core of Prosperity Home Mortgage, they are hyper-focused on communication, which includes everyone in the homebuying process. That includes the real estate agent, the lender, and the borrower. For President and...
The technology buzzwords for the last few years have been AI analytics, data analytics, and automation, but businesses are still unsure how to leverage those advancements in their systems. Leveraging data and connecting...
Medicare agents and agencies have gone through drastic changes over the last 20 years, making adapting and growing with the times challenging for all involved. With a constantly growing customer base, education and...