Every Client Is Fortune 500. What Do They Know About AI That You Don’t?

Jim Olsen: Like, and that led us to, moving into the space to create a company to do this because we ran into it again and again and again where they either had no process or they had so much process, it would take them twelve to sixteen months to get a model out the door.Host: Built this week, breaking it down. Built this week, we show you how. A fresh idea, a clever tweak, you locked in. Built this week. Hey, everyone, and welcome to Built This Week, the podcast where we share what we're building, how we're building it, and what it means for the world of AI and startups.Sam Nadler: I'm Sam Nadler, cofounder here at Rise Labs. And each and every week, I'm joined by my cohost, my business partner, friend, Jordan Metzner. How are doing today, Jordan?Jordan Metzner: Yo, Sam. How are you? Another exciting week. So much going on in all the AI laboratories, new products dropping, new real time models. So it's been another crazy week.Speaker 3: Stock market's crazy too. So just lots going on all around, but, super excited for today's guest, actually.Speaker 2: And with that, I'd like to introduce our guest, Jim, the CTO of ModelUp. And, Jim, if you don't mind, give us a brief intro of who you are, kinda where you came from, and then we'll get into a little bit of of seeing a demo of ModelUp.Speaker 1: Oh, sure. Yeah. I'm the chief technology officer of ModelUp. I'm one of the founders. Basically, where we came from was actually myself and the current CEO of Modelop.We're working in in the big data space before this and working with customers who are actually training AI models, that was over eight years ago. So working with real situations, and we saw a gap in the space, and that led us to actually developing this product, and we've been at it for over eight years now and evolving. As you said, it's constantly something new on the horizon. So evolving the product to keep up with the capabilities and what's actually out there being actually consumed within large enterprises. I know all of our clients are Fortune 500, so we're working with what's actually happening and being consumed within those organizations.Speaker 2: Amazing. Well, I would love to see a quick demo just so I can see kinda how the product works. But if you don't mind, Jordan's a more sophisticated listener, user than I am. Give me, like, if you could e l I five, like, explain like I'm five, the core use case for model op, and then let's see the demo.Speaker 1: Okay. Absolutely. The core usage is, basically, if you're going to bring an AI solution out to the public, especially as an enterprise, I mean, many of our customers are banks, you know, financial institutions, healthcare institutions, you better have some idea of the potential risks to your business, to your patients, to your customers, etcetera, out there. And we basically provide methodologies for how you assess that risk, how you do ongoing monitoring of the models themselves and how they're performing for your business, what are the costs of the business of that use case, you know, that's becoming a real concern now, understanding token usage. It really stacks up.So is it providing the value as opposed to the cost? And many of these different processes, and helping to automate that process and drive it forward so it doesn't become a burden on the organization, then they can take these great ideas out because it kinda sucks if it takes a year to get your idea out there. It's already expired in this space. So how do we accelerate that and get these AI solutions to market in a in a safe manner?Speaker 2: Amazing. I would love to see a demo. And while you're pulling something up, I... You know, my understanding is, you know, at least one feature of ModelOp is that you're... From the token usage use case, you're not only looking at token usage from the user, which I think, you know, a lot of us are trying to understand, but also from almost the business unit or what's it trying to achieve.Did did I understand that correctly?Speaker 1: Yes. That's correct. You know, we can look at there's the traditional view of looking at it from, like, for instance, the model consumption overall, but that's not really useful to the business or from the user. That's not really useful to the business because a user using Cloud Code to do something and develop some software is a very different use case than the user using the same model to perhaps, you know, answer customer questions or these kinds of things. So we need to actually understand what is the solution costing me.So for instance, like if I have a customer support chatbot that's out there, I wanna know what that particular business use case is costing me as a business on there, and also any potential, like, issues with it. So for instance, here we're integrating with Kong Gateway to pull in information about guardrails. It's imposed in blocks, specific requests on there. So understanding that over time as to how many exceptions you're getting, but then also what is the solution costing me. Because if it's just like a general HR query tool and it's costing me thousands per month, that might not be worth it.But if it's a a critical item like customer care support, etcetera, and it's saving new things, that's a very different cost equation. So being able to have that kind of information at your fingertips, so to speak, to to say, yes, this is worth the spend. Does that make sense?Speaker 2: It does. It does. And what am I looking at here?Speaker 1: So this is actually tracking a business use case, a hypothetical one here in this case, a customer support chatbot that, you know, our... We actually track all kinds of information about it, you know, the information about the business use case, but also, you know, potentially the implementation and stuff, which we'll show an example of in a second. But this is tracking the token usage of... And actually, in this case, imported from Kong Gateway. We integrate with them as a partner.We can also track it directly. We'll see that with some of our argentic use case examples so that we can see that that cost over time of of what it's actually providing as well as an overall dashboard of just, you know, all of the information about the different use cases, models, etcetera, and what they are are consuming in token so we can tie it all together and get that view.Speaker 2: Okay. And my follow-up question is, like, we've we've covered, token usage, and I and I think that makes a lot of sense on on how they're evaluating, you know, implementing or delivering, AI solutions to their customer base or even internally. You know, there's these other features, security, compliance. How does ModelOp help in those areas?Speaker 1: Yeah. Well, let me give you a a quick demonstration of how we drive that process that will kinda tie all that together. So for instance, what we're gonna do is we have these concepts of basically life cycles that are configurable, that can enforce different policies, but they drive the process through rather than being something that passively sits back. So we're gonna walk through a really, really simple example today where we actually use our discovery mechanism to query our enterprise platforms, find new models that aren't registered in the system yet, so called shadow AI. And then we're going to start a process to say, hey, we discovered one.These are what you got to do to make sure, you know, it's up to compliance standards in this case on that. So what we're going to start with is actually our discovery service where we're going to go out, we can query different platforms that are out there that are used. We're not interested particularly in desktop level because the reality is everything will have AI in the long run. You know, basically, Windows has AI built in. You use a particular API, it may have AI backing it.You don't know and etcetera. But what we really are interested in from our standpoint is the ability to actually discover, yeah, we've got these enterprise services such as Agent three sixty five, Amazon Quick, Amazon Bedrock, etcetera. What is actually being run out there? You know, Microsoft Copilots, there's people developing these within the enterprises all the time on their own, and we don't always know about it. We don't, you know, know what's being done, and that's a liability to the business because those enterprise level resources actually can have access to proprietary information.So we wanna make sure that they're so called brought into the fold, you know, automatically brought into the inventory, as I said. It's called the shadow AI problem. Some people look at it from a desktop perspective. We're looking at it from an enterprise resource level perspective of these real platforms that people are utilizing and utilizing a multiplicity of them. So it can be very hard to track down all of these different usages that are occurring out there.So what our system's doing right now is actually scanning for those and and reconciling it against known items within the inventory on that. And then what's gonna happen is when we bring them in, when we bring a new one in, we're gonna start this process where we go through and reach out to people to go get documentation, use our agents to auto complete forms based on the documentation they have, our own AI and agents, and then actually look at applicable policies based upon those answers on that kind of things on there. So now when we come back to the screen, we're gonna see it's actually found a ton on purpose. I didn't register a ton of stuff. So we found a ton of stuff sitting out there, all these Copilots, agent cores, you know, Amazon quick implementations, etcetera, that are out there.And we make it really easy to bring them into our system. For the demo's sake, we're just gonna bring in one today. We're gonna bring in a health care adviser chatbot that's written in Foundry. And for the demo's first perspective, we're gonna add a little tag on there, if I can type correctly today. There we go.And then it's this easy to bring into our inventory. So we're gonna do that, bring it into the system. And when we do that, that life cycle is actually gonna start. It's gonna start driving the process, moving it forward, helping us get it out there. So now it's telling us within the inventory, when we go back and look at it, and we look at our implementation, we're gonna see that Microsoft Foundry health care chatbot, and it's using g p d five mini, so it brought both in automatically into our system and bringing that.And you can see it's telling us we gotta upload some documents. So we'll go ahead and actually add the documentation to that. And then when we do that, if we can just add our use case, so we've got our use case form. This is just written plain English. And we get it in there.And then we go back to there and go ahead and approve and say, yes, we've added the documentation. Actually, approve that on that. Completed. There we go. See, it detected the fact that we actually uploaded the document.So now in the background, what's actually happening is it's going out and reading that document using our agents. We have a full agent workforce behind here that is implemented use... Utilizing MCP tools, a contextual database to actually store information about what's our... In our inventory, etcetera, so that we have all that information at hand. By doing this, we're going to rapidly accelerate the process.What it's going to do is this chatbot here, if I actually close it, has a full form that actually needs to be filled out. So you can see here where right now we don't have any of the form information. Well, the agent has actually just completed its work on that. So let me, sorry, close that down. You can see now that we actually go now.We've used the agent chatbot to do that. And you'll see now where it's actually filled out a Confuser configurable form with all the information for us, where it's going to be used, all of that kind of information done for us by the agents. So we're accelerating the process of getting this out here. And, you know, we can actually go ahead and move it forward. And when we do that, now it's gonna take those form submissions that were automatically there and go out and continue the process forward.And we're gonna see now where there's compliance violations, like we're comparing against the EU AI act and telling us what we need to do. So we can actually drive these processes and make it go forward, and I'm gonna end there so we don't have to go through all the form submissions, etcetera, and that kind of things. But I think you get the flavor where we're trying to make it so you can be compliant with all these regulations, all these things you have to do, your internal policies, etcetera, but making it painless. You know? That we just walked through what would take, honestly, months worth of work in some organizations to pull that together, and we can do it...Accelerate that into minutes.Speaker 2: Using AI to use AI, maybe.Speaker 1: Correct. Yeah. Helping to accelerate. Keep human in the loop still, but yes. And and Jim, like,Speaker 3: how how did you guys get into this? Kinda where did where did this thing start, and how did you get get it to where it's at today? Obviously, we've seen so much adoption and change in AI in and of itself. So how did it kinda... Where did you guys start with the product and and get it to where it's at now with kind of what we've seen in the market and whatnot.Speaker 1: Yeah. Well, as I mentioned, we were working... I was doing consulting at the time with some very large manufacturers, banks, etcetera, on that, and we were... It is in the day of the big data, you know, the data lakes, all that stuff, and that kind of things. And we were helping them to establish these kinds of huge data lakes for use specifically for training AI models.They wanted to basically, you know, one was a manufacturer, and there's hundreds of models they had that ran on their individual machines that were critical to their business. When working with one of those people, I actually ran into a situation where the only place where that model was being trained, that was being deployed into customer sites, was on a guy's desktop. He's just right there. If this if this person got hit by a bus, god forbid, tomorrow, they would have no idea how to reproduce it. They had no tracking on it.They didn't know what went in it. They had no versioning. And I saw this again and again and again where, you know, there there was just nothing behind it. Nobody had any insight into what was being deployed. So, you know, we we realized there needs to be a different process for that to actually track this information, validate the information, make sure you're compliant with any regulations, all this stuff.So we began to think about what that would look like, and that led us to moving into the space to create a company to do this because we ran into it again and again and again, where they either had no process or they had so much process, it would take them twelve to sixteen months to get a model out the door. So this became a real world situation when, you know, Fable rolled out and a bunch of different people started using it, all of sudden it was banned out out of the country. We had customers who now are scrambling to say, okay. Which solutions are are using Fable because it's now shut down? So everything it's using is broken.We have to we have to go roll this back to a different level of model. So by providing this kind of information about the actual implementations and being able to look in inverse and say, yeah, you know, Fable's being used by these 16 things, they are able to track it down very quickly. Likewise, you know, that's an extreme situation where a model is completely shut down. You can also imagine as, yeah, new model comes out. Well, who's using, you know, a GPT five mini, for example?You know? Oh, it's going into retirement. What? Who's using it? How do I upgrade?How do I ensure that it still works correctly? All of these kinds of things. So you have to know what you have, what is using what to even begin that exploration. It's super critical when we get into these rapidly evolving models. Okay.That's awesome.Speaker 4: If you're a founder or running a business, you already know how powerful video content can be for growing your presence. That's where video comes in. You just show up, share your perspective, and we handle the rest. HowSpeaker 3: can, people get a hold of you or find out more or learn more?Speaker 1: Modelop.com is, the easiest way. We have all of our information. We have demos that said you've only seen a tiny, tiny bit of the functionality. We have, more information up there and that kind of things as well as contact forms, etcetera.Speaker 3: Great. Sam, shall we do the news?Speaker 2: Let's do the news. This is, you know, kind of a silly one. Jim, Pet Libros, new AI powered feeder is a game changer for multi cat homes. I got a opportunity to, scan this really quick. I think it detects when your your food is low.See, it can detect, when your food is low. You know? Obviously, I think this is, probably helpful to a multi pet owner. You know? Do you haveSpeaker 1: pets I'veSpeaker 2: got poor cats. Buy... Oh, wow. You're the number one user here.Speaker 1: And trust me, my cats already let me know when the food's running low. That's not a that's not a big Yeah.Speaker 2: Don't need AI for that. Yeah. That's the question. Are are, you know, do are are there... Is everyone trying to force AI into everything when maybe it doesn't, you know, really need to exist?You know? I I have pets. It's it's it's pretty simple to know when to feed them. It's pretty routine. Secondly, they ask for it.What do you think of this product?Speaker 1: Well, as I said, it's... I haven't used the product. There may be advantages and disadvantages I'm not aware of, but, I mean, it does seem a little overkill to me for the price point ratio of, yeah, my cats will let me know. I can buy a a kind of a self feeding dish that just dumps the food out from gravity for about $10. So...And it's been working fine for us so far, but I can see maybe if you have specific dietary restrictions or needs or something like that. Perhaps something like this might be appropriate, but I would probably not be a consumer at this time.Speaker 2: Jordan, I know you're more of a dog guy, but I guess this would be relevant for... Jordan has three dogs, so that could be of... Obviously relevant for dogs as well. Would you buy something like this?Speaker 3: I mean, I guess, like, feeding the dogs is a daily routine. In my house, it's a twice daily routine times three dogs, so that's six times. Sure. I would love for the dogs to get fed automatically. Do automatic feeders work well with dogs?Not really. I think, like, Jim said, there are some, like, you know, kinda, like, you know, kinda gravity fed ones on the market. But but, yeah, I'm waiting for my Optimist robot to feed the dogs and to, you know, pick up the poop on the on the heart, so I don't think we're there.Speaker 1: There's a more useful. Can I get AI to do that?Speaker 3: That's a... That's my future plan.Speaker 2: This will be more exciting when it's the AI personal chef, and I don't have to think about preparing dinner for a family of four every evening. The one thing just to wrap up the show in one minute, Jim, I read that you're off... You're CTO. Yeah. You're off grid.You're Correct. Yeah. Give usSpeaker 1: a Yeah. So I'm, solar powered on that, and we're working over Starlink right now. So I have my own network that's powered up over here, and, I mean, it's continued. I've had it since the very beginning, and it's continued to improve. We get, like...I get, like, three to 400 down on that kind of thing, about 30 or so up. So as you can tell, it's working really good. It's actually a hailstorm out right now, and you haven't seen a glitch. So we're doing the... There.But, yeah, we're... Right now, I live at 10,200 feet in Colorado, and, yeah, it's... Power lines are miles away, and it's a dirt road to get up here. So it is pretty amazing that I can work from up here, and it's woody heat and all that kind of thing. Our own water out of a spring and the simple life.Speaker 3: Wow. That's incredible. So you have solar charging up batteries and then and then using the batteries in the nighttime type of thing?Speaker 1: Yep. Correct. So I have six different arrays of solar panels, and they go into lithium iron phosphate batteries, which is newer. I used to use lead acid, which wasn't nearly as good. These batteries are amazing.I charge great. They even heat themselves up if they get too cold. So, yeah, I run fully off of that. I do have a backup generator if needed. I have not run it in a long time.So, you know, we do generally produce enough, but when you get a long... Good bout of rain or snow or something, you might need that. So it's good to have.Speaker 3: Yeah. Like, with no sunlight per se, if, you know, smithers blocks the sun, how long can the batteries kinda power the house for?Speaker 1: Well, that'll depend on the load because, you know, I can check... I can obviously vary how much I turn on or turn off. But if normal load, I could go about six to seven days with absolutely zero sunlight Wow.Speaker 3: That's incredible.Speaker 1: I currently have. So I have eight very largeSpeaker 3: batteries. Awesome.Speaker 2: Oh, okay. Great. Jim, it was a pleasure. Thank you, everyone, and we'll see you next week.Speaker 3: Bye.