Episode Transcript
Hey, welcome to building the Loyalty Loop podcast. This is your host, Kyle James. Today we're gonna be discussing how an AI and information technology company was able to build over 50 large language model powered AI agent solutions for their clients in multiple industries, including the education space, using a customer first approach. If you're an IT company, then you need to listen closely. Have you found that your AI agent is hallucinating? Or just mishandling basic conversations. Speak to GPD Trainer. GPD Trainer fully manages your AI agent for you, eliminating hallucinations for good, while also distracting improving your AI agent's response. So if you would go to gpd trainer.com to learn more. Seriously. If you're an education institution, go to gp trainer.com to learn more. So for our guest today, we've got Hunter XO the show. Hunter is actually the founder of GPD trainer that has over eight years of software design consulting, generative AI and project management experience. and his team at GPD Trainer have successfully delivered over 50 of these large language modeled power adjacent solutions for their clients. Hey Hunter, welcome to the show. How are you doing? Great. Thank you so much, Kyle, for having me on the show. Doing well. How are you? Yeah, doing great, man. Just you know, beautiful day outside. No complaints on my end, so happy to, happy to be able to connect with you. I know we've been, we've been due for a podcast for a little bit, so I'm glad we're. We're here finally. So, but let's do this. So I know we got people listening in and they want to hear your perspective here. So building out these large language models for these clients, like AI is a big topic. Walk me through this, like kind of step by step, how did you get some of these top level education institutions and companies to scale their AI capabilities in this new era? So a very, very broad question, Kyle. I mean, I'll do my best to answer this one. So. Why don't we delve into one specific use case here. So you mentioned educational institution, right? So amidst a lot of the different types of commercial clients that GPT trainer currently handles, we do have some educational clients that, that stand out. I mean, I'm talking about large institutions, organizations. Ranging from universities all the way down to, you know, K 12 education in the us. The one I wanna highlight right now is tech, TEC, right? They're a European client specifically they are one of Denmark's largest vocational institutions. I think they have over 30 vocational programs. And then they, they offer a lot of continuing education and upskilling opportunities. Right. So lots of hands-on practical training. And they, they do face some challenges like, you know, the students not always being able to engage with the, the instructors on demand. You know, they all sometimes have day jobs, right? So scheduling is an issue. And also just availability of support, teachers limitations time-wise and sometimes even language constraints. Being the, the biggest challenges and for us here at GPT Trainer, we work hand in hand with this client to build out a custom solution at the beginning and then constantly maintain and improve it as time went on to deliver really good, consistent results that they're seeing benefits even to this day. So. talking about Yeah. like what were some of the different types of like results that you, that you saw them achieving when they implemented your AI solutions? Yeah. So for tech in particular I remember being told when we reached out that, you know, on average they're seeing for instructors that are using, leveraging our ai solution, saving about one to two hours per day Hmm. teacher, right? And that's, that, that really adds up. Yeah, I mean, it doesn't seem like a lot every day, but just from a, almost like a mental psychological perspective, there's a lot of busy work, tedious work that you no longer have to do, right? Because you can now, you can outsource that to AI on demand and have, it generates a quizzes that's based on your syllabus, your own reference materials for your particular class. And you don't have to write everything from scratch. And you could yeah. I mean that's just one example, right? And there's, there's always additional benefits where, you know, for students the challenge I mentioned earlier, not being able to access educational, human educational resources on demand, well now they can, there's an AI teaching assistant for that particular course behind the scenes, right? So they can go in 24 7 available. Can speak any language to accommodate the student's language of preference. And that's, yeah. So talking about those results that you, that you mentioned, like the one to two hours, like what do you feel like has been like the biggest, maybe like number one reason that you were able to get those results for that, that that ai for the AI within the education space. So here I think it's really good that I mean, first of all, you know, it's a really good question Ka because, the, the, the word chat, GPT comes to mind. You know, teachers, why can't they just go type in chat GPT, you know, so something like, generate a quiz for me. Well, the thing is people tried, our clients in a similar position have tried that in the past, and it just doesn't work as well as they expect. The thing is chat, GPT is trained on public data, whereas, you know, the solution we built is anchored on the body of knowledge that the client provides themselves. And on top of that, we are a service based agency. And so we don't, we don't just give, you know, clients access to software and then walk away. Ever since they engage with us, we are on, you know, regular check-ins with them. In fact, we have sending meetings. We try to ensure that whatever solution we built is constantly being monitored and upgraded, right? Over time based on the changing demands. Fortunately for tech, you know, for the most part. You know, after we spent a couple weeks configuring, optimizing the prompt, the training data, everything's been working out pretty well. But you know, there are advancements such as when Deepsea came out you know, deep seek. Mm-hmm. you know, it's an improvement over both cost and performance. And since tech is not super sensitive about. You know, obviously data being shared with the language model providers, them being an educational institution and all we were able to leverage some of the more advanced, more powerful models to further improve the performance of their, their AI agents. And just to get, they're able to have access to that, that same model of deep seek. Right. And CLA, philanthropic or anything else? Is that what I'm hearing? absolutely, I mean whenever there's a powerful model coming out in the market. You know, GPT trainer would evaluate evaluate, its, its competitiveness, its performance internally, and if it hits the benchmarks, we actually introduce that into our framework. And so our clients will automatically enjoy these updates without having to worry about, you know, building out the integration themselves. All they have to do is pick from the dropdown menu, select the, the option, and, and that's it. Yeah. And that's what we do too. all hosted in one. Like they don't have to go in, I just, they keep, keeps it all within GPD trainer and then they Yeah, click the one they want and then boom, plug it in. exactly. You know, GBD trainer is designed to be low code or no code out to, to work out of the box for clients and for tech in particular, right? We have our own AI engineers helping them configure things so. sure. In some ways they don't even need to log in. You know, we, we take care of that for them and just send them the link to the bots and they distribute it for usage. So looking at like kind of this, this, this you know, student support or, or not even student support, like customer first or student first support. Right. What have you done or GP, what have they have done over the last maybe like six months to improve the delivery for their clients? So over the last six months, right, from a platform perspective, we've implemented a lot of new technologies or evolutions in terms of our, our proprietary AI supervisor and multi-agent framework to really refine the accuracy behind, say reference materials that's being used to anchor the training context. For these chat bots we have, you know, improved the intent classification. Meaning if you have multiple agents behind the scenes, I, I don't wanna get too technical, but in general, right? The supervisor decides which agent for the chat bot, which AI agent, for the chat bot is best suited to answer the user's query, and then routes to that agent, right? So the whole routing workflow was update upgraded. We also have. Variables that you, that we use to monitor the conversation, or I should say background agents with variables. And we capture key ideas into those variables. So you are turning unstructured conversational exchange into structured information that you can then forward to a database that you can reference later. And this, this, this whole thing forms a feedback loop, right Kyle? So, Okay. You talk about continuous learning of AI models, you talk about self-learning AI agents, and this lays the groundwork for that as you capture and standardize information during conversation, conversational exchanges. Anything that's good can be used to reinforce the chat bot when it answers future questions of similar nature. Anything that's wrong, you can disregard, you can tell the ai, don't use this answer, use the revised version. Right? And over time, the the AI gets better and better. And so, you know, the longer you stick with the platform, the smarter the agents become. And you know, it, it's, it's a self-improving cycle. Wow, incredible. So I see that so much in like this market of the AI specifically. I just feel like it's constantly changing and, and shifting and, and getting, you know, more effective and faster. But like for, for gp, you're trying to, going back to to, to your, your company like, like as far as like the next six months go, like, not the previous but the next six months, do you feel like your team is gonna be focusing on the most moving forward, especially in this changing environment? Oh yeah, absolutely. I mean the first thought that comes to mind is certainly at the most foundational level, there's gonna be new language models. I mean, this whole, you know foundational language models, I. Seeing the landscape is evolving very quickly. Just last Saturday, over the weekend LAMA four came out from Meta, right? That's a new model that we've yet to, to test out. If it works well, we'll probably introduce that in our list of supportive models. And I'm sure in the next six months, OpenAI is gonna have new models. Claude, Google, right? They're gonna come out with new models. Even deep seek probably is gonna come out with new models and we're gonna. Be on the forefront monitoring as new language models get released. So that's the first thing we'll improve. The next thing we have in mind is this concept of chaining agents. So chaining, right? Chaining together multiple agents. So right now, GPT trainers multi-agent framework works in parallel. So the AI supervisor first routes the user's query based on detected intent, right? Two, it's designated agent of choice, and then that agent handles the user's input. Well, it's one agent among many that handles it. What we wanna do is implement something that's similar to language models, chain of thought capabilities in reasoning models, right? Like R one, like O one in GP Chat, GPT, and in oh three Mini. But now we're trying to make it baked into our internal framework, so. You can chain agents together using any language model, and you're doing it sequentially, which means you are much more capable of advanced reasoning. And, and you do the multi-step thinking every step of the way. And you have full control over it. And at every stage you can even consult external information through function calling. You can. That's right. So, you know, extract real time data by calling an external API and having the response fed back into one of the chain agents. So things like that. Again, I don't wanna turn this into a very technical discussion, that's great. So it but like y'all, your team is like taking it to the next Absolutely. updates happen. I. You guys are right on, you're right behind it. Looking to implement it for not just your own right, for the company, but also for your clients and customers are really just trying to stay on. They're trying to stay on track with what's happening in the AI industry. I mean, it's That's right, Kyle. I mean, yeah, yeah, definitely. That, you know, 100%. So the, the new capabilities that we have in mind will most certainly open up new use cases of GPT trainer that perhaps was a bit of, you know, out of scope before. I'm talking about, you know, Hmm. things that are more reasoning focused, right? Maybe even long form text generation that require iterative refining. Right of the content. And eventually even doing simple math. I mean, gasp, right? Usually language models aren't known for math, but perhaps with this chaining approach, with function calling built in, it can do some, some mathematical analysis even. So a lot on the horizon that we have in mind and should really, you know, bolster the platform's capabilities over time. Man. That's great. I love it. And Hunter just for, for everyone listening in on this, on this conversation today, where, where can people learn a little bit more about you and about G PT Trainer and their services? So I'm generally very open. I always respond to messages through my personal email. And also we have a Discord community where you can find me there pretty much all the time, right? Besides when I'm sleeping. I, you know, I usually respond to messages whenever I have you know, whenever I'm not in meetings and when I see it. There's also, you know, if, if you wanna schedule in advance, book a time on my calendar just email me ahead of time and I'd love to, to have a chat explore kind of AI use cases with you. And, you know, that's generally my attitude. Cool. Awesome. That's amazing. And so what we'll do is we're gonna wrap it up from here. Thanks everybody for listening in. Thank you so much, hunter, for honestly joining us and sharing your insight on how to be a trainer has really helped, which is incredible. Over 50, 50 different educational institutions and companies by deploying large language models, powered by AI agents that have changed the way schools and businesses have operated, saving them hundreds of hours per year. And if you're an IT company out there. Your AI agent is hallucinating or just mishandling basic conversations. Again, go to gpd trainer.com to learn more on how we've helped our clients fine tune their AI agents and maximize their time. So again, thanks guys for listening in. Hope you have a wonderful rest of your day. Looking forward to seeing you on the next episode of Building the Loyalty Loop. and take care.
