Awesome. Thank you. Well, I'm amazed to see the standing room only and I'm excited to dive into what we're going to showcase today because it's kind of an exciting time. I just wanted to start by saying we were making rocks think. And like we're getting to a place where those rocks and those things that we mold into a certain shape and size are starting to transform how we think as humans and what our economy looks like. And it's just me being at this conference and I know our team being at the conference here, we're learning right along with you all. It changes every day and just exciting time to be in the space. And the fact that you're here and listening is reassuring that we're all learning at the same time and we're all kind of ingesting what everybody else is doing. So with that, we'll kind of jump into the next slide here, maybe. There we go. So what we're going to go over today, there might be a little bit of an introduction, and I'm going to talk through some of the things that we're doing at Running Robots. But I think that the concepts that we're outlining and the things that we're talking about can be applied in any business. We work pretty heavily with manufacturing businesses. And when you're hearing our talk around, let's say, SEO or pay-per-click or website, I want you to equate those things in your mind to the departments within your own business, regardless of what that is. We'll go through the structure of our data, and then we'll get into a demo, and Megan's going to help me. There might be some crowd interaction. with some QR codes and some missions that you guys can give us. If we end up using your question, you'll get a running robots mug from there. So try and think of the best question and we might use yours in the demo here. So with that being said, the framework we're going to go over today and what we're highlighting is really something that you've already technically been using if you've used the CIRAS AI website. As Paul stated in the intro this morning, we came to him and said, okay, you know, we want to build the website, we want to make it AI first. And how do we make an AI website and make it so that you all can use the AI tools that you have on your machines? and be able to get the most out of the conference. And I think that the framework and the data that we put together on the site is a direct relation, indirect relationship to what we're going to be going over today and how we operate as a business. So in the future, as these AI agents evolve and as they scrape the internet, as they're going out and starting to understand the context of the internet and what we've put out there, we believe your business is actually going to have two websites. One is the human-based website. You all know what that looks like. You all know what that feels like. You're seeing it with your eyes, you're navigating it with your mouse, and it's great. But the second website is a new concept for a lot of people, and we just wanted to kind of highlight what that is and why that's important. So the image on the right, I know you can't read the text on it, but it's basically a, let's say, an MD file or a code file, something that actually can be ingested by the agent. And the cleaner and more efficient that you make that file and the more that you add to those explanations for the agent, the more that you're going to actually be seen within search results or prompts within the generations of those LLMs. So the two website scenario is something that we want to kind of highlight here. But it does tie directly into that structured data approach. So as I said a minute ago, we're going to talk about marketing lingo. We're going to talk about SEO. We're going to talk about pay-per-click. But try and equate those things in your mind to the actual departments within your business. Think about shipping. Think about welding. Think about the things that you do within your business that could actually equate to customer success and customer knowledge. and how you communicate with your customer could actually use these same same concepts going forward. So, what is it? What's the process? You know, I named the slide or the PowerPoint around, you know, the broken process. What's not working? And one of the things that we see from that is not looking at things from an AI-first perspective. Understanding, hey, there's an actual LLM, there's a bot, there's a thing that can help translate what it is that you do in a much more efficient way than you're probably doing today with your humans. And being able to relate that information down to the human, to have them understand, or to, let's say, the example I like to use is actually when, you know, we all know the person that really only responds to texts. Maybe they don't have an email. Maybe they don't have, you know, an inbox that they check regularly like I do. So it's, how do you actually put in your CRM or put into your systems that Johnny at XYZ company really only responds to a text message or wants to receive these updates via text. If you're able to, let's say, meet the customer where they're at, that's where we think the power of this framework and this system is really going to demonstrate today. I was going to say too that it also lives in three areas. So like when a vendor has information around, okay, how the product should be configured for your business, that's all in their head or all in that type of system. The dashboard, I'm guilty of this. We send out, you know, monthly reports. and we look at the metrics of how people are using the dashboards that they said, hey, they want to create. A lot of times they're looking and they're scrolling, but they're not really ingesting that information. It's not really relevant to them at that significant moment where they maybe have a question about how to pivot on their business strategy with just the information from the dashboard. And then the AI tools, as good as they are, they still hallucinate. So how do we help the system or help structure the system so that the hallucinations are mitigated down to as little as possible? There will still be hallucinations and there will still be things that are not accurate, and they are not to the point now where we're like, just ship it out the door without human oversight. There still has to be that human in the loop scenario. So the three principles of how we reorganize as running robots and what we do when we look at the systems that we're using or running today is really before we automate, we start to look at every task. And Megan can probably attest to this firsthand and the amount of hours that we've spent into our project management system. and just going through the list and saying, does this apply in the age of AI? Does the customer get benefit from this? And how do we understand what's the output of that task or of that sequence of tasks as they fit together within the system? The other thing is AI inside the system, and using AI for what it's good for, but not just bolting it on the side and saying, oh, by the way, we're AI first because we put a chat bot on our website. That's a different scenario, and that's a different way of operating with these new tools. The memory aspect, the third principle, is that without that additional context of what happened last month, This month doesn't really mean much. So by actually having a log or a system that can actually go through version control of what you've changed and how things look, that system doesn't really have as much value if it doesn't have context of where it's coming from and potentially going to. And so that memory layer We're playing around with a solution called Mem Palace. I don't know, has anybody in the room heard of Mem Palace? Okay, I got a couple geeks in the room. Cool, cool. But the idea there is that the actual information that's stored within a palace and a room or a hallway is really adding that structure to the data. and being able to understand, okay, what happened last month and what happened this month are different pieces of data, but comparing them in a particular way allows that data to be ingested and used by the LLM much more effectively. All right, so here's where we get into the good stuff. And this is where I had a little fun with this presentation. And today we're actually kind of announcing or releasing our version or our remade architecture as a business and how we're actually telling the world of how we're going to interact with our clients. So our clients haven't seen this. but we're really actually starting to roll this out here in the next month or two. So every client and every business, every person that we interact with as a business is actually going to get a robot or an avatar. And within that avatar, and why that's important, is that we're actually assigning properties and attributes to the things that they're doing with our business, but also in how they interact with our business and where they're at in their own marketing knowledge. So we're meeting the customer where they're at and understanding that, okay, maybe Johnny with just a cell phone and not an inbox is going to prefer that text and that information would be stored within their Avatar. So let's look at the pieces of what we've built. Again, translate this to your business, because I know you're probably not, you know, let's say, going through the automation integration or content creation pieces, but translate that into something that relies on your business. But for us, we've assigned different body parts of the robot to the different service departments that we operate. And why do we do that? Why is that important? Or how does that actually relate? Well, it makes sense when you see this type of robot. And yes, there can be a one-leg, one-arm robot because it's instantly visual and we instantly see, okay, maybe this robot isn't running as fast as what it could be. with these other legs or arms of the service. So by having that visual and having that understanding from a human perspective, we can instantly see, okay, there's things that might need to be tweaked or updated within that client's profile to give that robot a full avatar. This is where, I guess, the fun comes in. So one of the things that we struggle with as a business is that when we come into a client meeting on a monthly basis, there's a lot of, let's say, what did we talk about last month? Or how did we, what did we actually go over? Or what's the next step? What are the different areas of our business that we can improve? And there's a way in which we can architect that, but I thought, why not make it look like a monopoly board and have the monopoly board actually represent the different parts of our business? And that actually helps our customers understand where they're at and how those different services relate together. And then when they're actually interacting with the business or coming to their dashboard, they can see this and click into each one of these boxes. So we've got the four different departments or the four different areas. One thing I would call out is this master data download. And what that does, it's kind of like what's on the CIRAS website. that we've introduced here is that you can really get all of the structured data from the different departments into your LLM by just going to that download. But what happens when you click, let's say, into one of those boxes? It actually allows us to focus that information and get a little bit more granular in the operation and what they're actually doing with that data. So you can get the overall, let's say, full marketing context if you click download data on the Monopoly board. But the individualized data per department actually allows us to say, okay, I'm going to ask a specific question around And this is AI visibility. So I basically clicked into that little box. But from there, you can see the two scores. And what are those two scores? One is the knowledge score of the customer. So in our meetings with our customers, we're recording them. We're understanding what words, acronyms, things that they're using. And by doing that, we can assign them a score to meet them at their level. So if somebody comes in and starts talking about ROAS and cost per click, and they're getting down into the nitty-gritty of marketing terms, by doing, assigning the score, we can actually meet them to where they are. If they're just saying, hey, I don't really know this this pay-per-click thing, help me explain how Google Ads works and all of that. We can meet them. Their score might get a little lower, but at that time, it'll actually help them understand the concepts and bring their score up over time. So it helps our team and the bots understand where that customer is when we're delivering the message. Come. One thing with this slide as well is that we have an ability to, let's say, chat against the data and use our bots. So we're using our system prompts and our information, which I think is useful, but it's not necessarily an AI first when it's not the customer's information. I feel like getting the information into the customer system on their desktop, laptop, whatever it is, and allowing their bots, the ones that are trained to communicate to them, where the value is added, if that makes sense. So the scores, and I just went over this a little bit, but I want to go a little bit deeper. And the terminology and things that we do within the score can increase over time and help that robot go from that small little robot that you saw in the beginning to a bigger robot in the end. So by the size of the robot, we're actually determining The score, and that's another visual cue for our team to understand where that person or business is in their marketing journey. So what we're covering here, what the framework is that I'm outlining here is really from avatar of meeting the person, understanding who they are, getting into the game board and explaining, okay, here is how we operate. Here is the process that we go through on a monthly basis. And then we're actually looking at the individual departments. So clicking into one of those squares. seeing the data that's there, and understanding that the information that's presented there is not necessarily created just by AI. It's not just a box and data filled with an AI output. It's actually our team's input that created that dashboard. So each department head of our business is actually critiquing and understanding the flow for that customer and getting information. And I'm going to go through that here in a second. But it's essentially not using AI up until the point where it actually gets the information from that system. I didn't show the Plinko board. The Plinko board is still a concept in play, but I wanted to address it in that the game of Plinko or the Plinko operation is, you know, you drop a ball or a token into the top and it hits a bunch of different pegs on the way down, and then it ends up in a certain area. And by visualizing in that way and saying, okay, a customer came in from pay-per-click, they dropped the token in and they hit the homepage, they hit the product page, they hit the product category page, then they hit the contact or the checkout. And by visualizing those things as we're going through the data and having that visualization, through the marketing process. It helps the end user understand where those leads came from, where those pieces of information are being used, and just again, adding that visualization that the customer can then say, oh, okay, aha, I see where that information is coming from and how This data was put together by the robot team. So what you take home. So one of the things that we're doing here, and so if you can see on the right-hand side, we've got a GA4 export, a Search Console CSV, a Screaming Frog Excel spreadsheet. These are all pieces of data that we're capturing from different services. And so our team has worked together to restructure how we operate. And I don't know if anyone was in the room the last year when I gave a presentation that was all about N8N and how we connected Google Ads to your CRM, to your ERP, and then pulled all that information down to make a really cool report. We're still all in on that piece of software, but that's not really the key takeaway here. It's basically that AI is great for ingesting and understanding data, but we don't want to introduce hallucinations into the ingestion of information. So we're still using tools to go out and grab that information from Google Analytics, from Google Ads, from Screaming Frog. We're scraping raw data so that we have an understanding of where that information is, and we know that it's not hallucinated against. So there's a place and time to use AI, but then there's a place and time to use AI to actually just get the information Directly from an API call or directly from a dashboard or a system that's not using hallucination type devices. And that's really leading into the trust by design. So by doing this, and I'll go through this real quick, the clean APIs, and I just listed those different pieces or those different tools, the AI's job is small in that regard. It adds a small bit of context to say, hey, here's what I saw in the data, maybe in comparison to last month, but it's not actually manipulating that information in any way. So it's kind of a separation layer or a way to actually say that we trust the information that we're putting within these files because we know AI hasn't touched it yet. There's a summary, but there's not actually manipulation Or getting that information directly from an LLM. The other thing here is 1 file per department. So when we first started doing this, I was in a talk last year, and the person giving the talk, I won't say the name, but they basically said, hey, I took all of my information and just shoved it into the system. I took our HR manual, I took our SOP, I took all of our stuff and just put it in there and I started asking the question. It was great. Well, if you do that, the context window is muddy. It gets really inaccurate, and the AI doesn't really know where to look or what to look at. So we've spent a lot of time as a business trying to understand that context window and how to organize the marketing data within that context window. So we get the best answers out of those things. And so by creating the basically MD or markdown file per department, we're then able to structure that data or point to the information and the device or the LLM can then articulate or say, aha, here's a good summary of what's there. But not so much that it's just overwhelmed with information and it starts hallucinating, if that makes sense. So what happens through this process, or what are we going to see in the demo, and how do we actually understand what's happening? It's one, pulling the information from a system, as I just mentioned, one of those many logos on the last slide. We're generating a short summary. A person from our team once a month or once a week, will be reviewing that information. So that lives on GitHub. We pull that information over, and the person or department head at Running Robots looks at each client's file. Maybe they're just looking at, maybe they're the pay-per-click specialist, and they're just looking at pay-per-click data. They're opening up that file. They're reviewing the summary, they're reviewing the information, and verifying as a human, hey, this is actually accurate, or there's something wrong here, maybe we need to tweak how this works, or something that needs to pivot within the account to make an optimization. So there is human oversight and review in that. in that workflow. Once we actually say, okay, yes, that information looks good, we like the output, we like the data, we'll do a commit. And that's just basically a geek's term for basically pushing something to GitHub to say, I commit this to a public or private repo to say, here is the information. and how we're going to share that to a client. So once we hit commit, that information becomes live. And what I think is the power there, or the reason why we're actually pivoting towards GitHub instead of a monthly email with a PDF attached, is that's instant. So as soon as that person hits commit, as soon as one member of our team says, okay, this file looks good. That information is then live for that customer. They can go and prompt against that repo and say, what's new? What's happening? And it just really depends upon the cycle of time that that customer wants to review that information. So if we're, let's say, every day reviewing that client, Maybe it's a new product launch and we want to look at that data every day. There can be a commit or a cycle that couldn't be available prior to using these tools. And so our speed in which we deliver information to a client and the speed in which they ingest it has drastically increased and the ability for us to really help them make business decisions, even a day or two after a product launch, makes total sense. And it really speeds up the ingestion or the, let's say, usefulness of that information. The sharing part is really just saying, hey, this information's out there. And by allowing others within your organization, maybe it's not just the marketing director that gets this, maybe it's the CEO, maybe it's the product specialist that can all have access to these files. They can ask different questions and have different avatars around that same data. So if your CEO might not, you might think he knows a lot about the way marketing works, but the chief marketing officer really realizes that maybe he doesn't know what he's talking about, we can tailor the responses or tailor the avatar of that response per business role and per customer. So that really helps in how we're actually ingesting that and how the, let's say, CEO perceives the data versus the marketing director. So growing the knowledge base and why do we use GitHub or what is in this GitHub thing that you keep talking about, Adam? Well, the information is organized in a very simple way. Who in here is actually used or know what I mean when I say MD file or markdown file? Okay, we got a couple. We got geeks in progress here. So we've got a couple of geeks in the back, but we're learning Markdown. And if you think of Markdown, if you look at it that way, it's really a fancy text file. Honestly, that's all it is. It's a text file that just has certain ways of formatting information that make it easy for LLMs and bots to understand. And so what we're doing as a business is taking that information, making it a markdown file, and putting it in the structure that you see here. So as a repository goes, and we'll kind of open that up here in a second, but we'll actually let you guys see the public repository, see our new structure and actually prompt against it with a demo client that you guys will see here in a second. So if you want to see the repo and what we've put together or how we structure that data, please just scan the QR code. This will take you right to the public repo and we'll be actually asking questions against it in our demo. So, you'll kind of understand who that is. Everybody got the UR code? I see the phones are still out, so I'll just hold on here for a second. All right. So next, what are we going to do? What's the demo? What is the special sauce? How are we going to, you know, demo this lie? And what we're going to do, we used fictional data, and I'll let me back up here. So I started by asking a client if I could use their data in this presentation. And I started basically saying, okay, we'll take the data, I built the repo, and I said, okay, tell me about this client. And I asked it a couple questions. And after the first two questions, I was like, I can't use the data. I can't put this information up on the screen and ask customers to ask questions about a live customer because it was too good. It was actually beyond the place where I was like, okay, if there's a competitor in the market, it's just going to be an instant, okay, here's what they're doing as a business. So we did make up a client, but we used the structured data. around that client to actually articulate what's in here. So the information that you're seeing and the data and the responses is accurate as far as how a typical business information suite would be ingested into this framework. So, and actually let me back up real quick here. It's Prairie Ridge Manufacturing doesn't exist, made the logo with ChatGPT, but we did give it a name. So it's manufacturer is the made-up person. So Manny is the president of Prairie Ridge Manufacturing. And he is a relatively, let's say, not advanced, but he is, let's say, not a beginner in the world of digital marketing. So he's one of those that would come in and say, you know, what is the ROAS of the pay-per-click campaign? What is happening within the marketing of my business? And they're not necessarily a newbie when it comes to talking lingo or marketing speak. But we'll kind of show you a little bit of a demo around how the information changes between Manny and what was the other name? It was Algo Rhythm. Algo Rhythm is the We'll see what Al's response is here. But one of the things that we want to do first is, so I'm going to go to the next slide here, and I might go back and forth. But if you can think of questions that you might want to ask Manny's bot, so if you put yourself in Manny's shoes, You can scan this QR code and email Megan questions. So if you scan that, it'll pull up your email, and then you can just type in a question, hit send, and then Megan will be sitting here watching the inbox, and she'll feed me links while she's prompting, and we'll be able to see some of the Claude responses. So And if we pick your question, you win a mug. So please type away and let us know if there's any questions that you would like to ask against this repo. So let me actually back up here real quick. Is that one of the things that we're demoing here is that the information on the GitHub repo and how we actually operate the business internally would be a two-way st. So while you're typing these questions, I just wanted to say that when you're asking your bot questions, let's say that you're asking, hey, how does this how did we do last month? Or how did this product launch go? If the answer comes back, and let's say it's not producing the information you would like, we can have, you can ask your bot, hey, since you have rights to this repo, please ask the robots to add data around this topic. and maybe the answer wasn't what you would anticipate getting back. If you're asking your bot to say, hey, ask running robots a question, it's feeding information back up into the GitHub repo that will allow our bots to ingest that information and say, hey, Manny asked a question that really didn't get answered very well, or maybe we didn't have data around that. that information, next month or next week, we would add information or work with Manny to get access, maybe it's about their ERP or other things within their business that maybe we don't have access to within the repo. By asking that type of question and saying, hey, here's what I wanted to see, here's what I didn't see, We're able to get that feedback, complete that loop, add the data, and then next report or next time that question is asked, that repo is then updated. So it's a very fast, continuous feedback loop for how you're interacting with the business and how the business ingests the data to make educated discussions. So, with that being said... We have a lot of questions coming in. So we've got our first one that's currently running, but if you want to... Okay, yeah, I'll go ahead and share an example real quick here. And on the bottom of this, it's may the prompting gods be with us. So whatever comes out of this, we'll see here. So let me just change the display here. So, I'm going to mirror what I've got on my screen. Okay. So this is a predefined example. We get kind of asked these questions to start while we're getting questions from customers. But essentially what we've got here is a, you know, please answer the question using the following data with just a straight link to the GitHub. repo, and then how did we do last month? That's all they typed in. That's it. But with that, the bot is actually understanding the context because it knows, hey, Manny is asking the question. Manny is an advanced marketing data user. And we got this information back. So we're basically looking at You know, the headline matrix of the number of sessions, the RFQs. We can really kind of go down and I can go deep into the data here. But essentially what we're getting is a top to bottom understanding from, let's say, Manny's bots perspective, how they did last month. What are the things that maybe they asked in the previous months that caused this data to come up? So it's an ever-evolving way of communication. So let's look at, so this is the advanced marketing background bot. But let's look at the, maybe the CEO that thinks they know marketing, but they really don't. The, let me see here. So this would be the algorithm. And so we're getting some data here. So it's, you know, maybe a little bit more streamlined as far as how it's articulating the information. But if we look at the top line snapshot between the two, so if I just go kind of, I'm going to go back and forth here. You know, we've got a lot of acronym soup on the advanced side, where if we switch over to the... Pay-per-click. It's, again, bringing that information down to the level of where that makes sense for that person. You guys see the difference in how those responses work? Same data, just different profile, different person, different response. OK. Now, let me pull up, center first five one, OK, which is from... So, which channels saw the biggest increases/decreases month over month, and why? So direct fetching, before I dig in, got context, Prairie Ridge manufacturing, demo client, channel data lives in websites. Let me check there. April is the most recent. I'll pull April plus March, month over month comparison. So let me open up the MD file here. And we can just take a look at what we got back for that question. So very rich manufacturing, month over month, April versus March source. And so we can see here, the biggest gainer was organic search with a 24% increase. So I don't know who asked that question, but you can see how that simple Let's say, month-over-month change was highlighted instantly within the results, and it's, let's say, articulating that answer really well. We're working on the second one here. Working on the second one? Okay. Let me go into one of these other ones here. So the, you know, which pay-per-click campaign is wasting the most money. So this is again into the Manny side of the scenario, but this is highlighting, okay, going into that specific MD file around pay-per-click marketing. And so it's not necessarily scanning or let's say looking at the other areas of the repo because of the way we've designed it. It's specifically looking at the pay-per-click marketing MD file and what's contained within that. So the response, PPC waste analysis, verdict competitor tier one fabrication was wasting the most money. So we've got spend, clicks, leads, CBR, and CPM. So it's worst offender, the account average, 24, let's see here. So, a 5.2% bounce rate on landing page rate. So, you guys can really understand or start to see the granularity that you can get into on a particular department or as broad as how did we do last month. That's a very different response or a different approach. OK, got 10 minutes left. Got it. OK, we have another one. I just sent you one more. Just sent one more. OK. See if I can whip one more up here. Let's go with that one. Okay. How do we standardize our components with our custom engineering engineered customer orders? This is a good one. I like this. Who did this one? Okay. Awesome. That's a very, we get this question a lot and we have a lot of customers that are very, let's say, customized in how they're delivering product. And one of the things that we struggle with from an e-commerce perspective is, okay, if you customize every order, your SKU numbers are infinite. and the ability for a bot to understand your information and how you operate as a business, adding that context into marketing and adding that context from your ERP. So like in our previous talk last year, I was talking about, okay, connect your CRM to your ERP to your website and have that information flow effectively. And for somebody like yourself that has, let's say, customized parts across the board, it becomes difficult for that information to fit within a bot context window. And you get a lot of hallucinations when you get over a large number of product or how to, let's say, this component fits with this component fits with that component. And by segmenting the information, understanding how the maybe components are fit together. So in this case, we would potentially build out the repo so that maybe we had attributes of those products. Instead of trying to list every SKU known to man, it would be, okay, these are the different parts. of their manufacturing process. Here's how their business works. And we can specialize the responses or specialize the marketing approach based on those different attributes. So it's not always just, here's my product list, ingest the product list, here's my marketing data, how did we do? It's really, how does this specific instance for this business equate to a LLM's context window. So I'm interested to see what comes back here. So let me pull up the MD file here. So this is a good one. So right now, we do not have a documented component standardization program repo. So it's actually calling out, and this is one of the things that we put within the repo, that we help the bot not hallucinate. I'm not saying that we've solved it, like, okay, we've made it so that bots don't hallucinate. They still will. And we have to cross-check this information. But One of the things that we try and do in this structured data is state that if the information isn't within the repo. Don't answer the question. Don't try and philosophize on how they could do it or pull numbers out of thin air, which they do, but we're trying to steer them in a direction that says, the customer asked a question that didn't exist within the repo. Would you like me to flag that to go back? to running robots to say, hey, maybe we're incomplete on our ERP data. Let's pull that in for the next month's report or the next repo commit. And then we can ask that same question again and see how those responses compare and see if we're asking the right questions. Okay, before we go to the next one, does anybody have any questions that they would like to ask to us instead of just our bots or other things? No? Okay. Yeah, go ahead. I teach engineering instruction for like a couple of years at Des Moines Area Community College before they transfer to Iowa State. And so I'm here to kind of figure out how to help my students be prepared for the workforce where AI is going to be prevalent. So From an engineering perspective, which tasks are they going to be able to perform that are kind of AI agent ready and which ones need the most human oversight at this point in time? That's a really good question. I think that the best way that I've found to prepare someone for using AI is to use AI. It's a very simple answer to that question, but in your interactions with the information, so one, giving it and knowing what structured data looks like, and to understand what context you're giving the bot is, let's say, one aspect of that. But then asking it the different questions or different phrases of that question, even just changing little bits and pieces of that question, you start to learn how to communicate to that thing. or to that entity to get the output that you desire. So using AI, the only way to get better at it, in my opinion, is to use AI and continue to ask questions. The cool part about it is what we're doing here is that I don't feel dumb when I'm asking my bot to explain something to me that I don't know, whereas I would a human. So the barrier to feeling stupid has been dropped by using these things. As I was preparing for this presentation, I just have to share one quick story. Like, we're pouring a new concrete pad in our front yard, and I couldn't really spend all the time I wanted to to prepare for the presentation this weekend. So I put my headphones on and just started asking my bot about my presentation. My neighbors thought I was crazy, just talking, like going through my presentation in my head and, you know, carrying cement bags. But that use and that use case of basically just putting your headphones in, turning on Claude, saying, here's my presentation, help me walk through it. that use case of learning with AI and just stepping through the process is, I think, the best way to learn and how these, let's say, articulate or how they work together with us. Yeah. I got one. So you talk about storing a lot of vendor knowledge, you know, all your client knowledge and all that kind of stuff. There was a comment earlier in the opening about tribal knowledge. How do you, have you found a good way of going through and getting stuff that's not already on a piece of paper that lives in people's heads? That's, again, another really good question. What I would say there is, as an organization, one of the things that we do internally is every Friday, we're meeting as a team and going through the use cases of how we're using AI. And one of the things that we're stressing in those group sessions or those work sessions is looking at your daily output and saying, okay, what is it that you've done in the last week? What are the things that you're doing? What are the projects you're working on? And asking our bots to help us understand what commits they've made. So by, and I sound like a GitHub salesman up here, I apologize, but by using GitHub and allowing us to basically say, Tiffany made 25 commits last week and she worked on these seven projects by understanding how they're communicating with the bot or they're communicating with our customers and having that information within a repo, as an organization, we're starting to look deeper into those individual roles and get that context maybe out of our brains and into a written form that we can ingest and see the trend over time. So to answer your question shortly, GitHub has helped us see that version control. And by adding things that you typically wouldn't say are code that needs to be on GitHub into those repositories for our clients or for our employees has really helped us start to get into the granularity of how they're working. How do you scale into a business if you're going to introduce yourself to a company that hasn't embraced this? Good question. So the onboarding piece is really one, as soon as we meet you, we're going to be creating an avatar. We're going to be looking at you, looking at your knowledge, looking at the business. and starting to grab at the different pieces that you use to operate that business. So your CRM, your ERP, you know, your inbox, maybe your chat history, just all of the things that we can actually pull from you that you feel comfortable sharing would be basically a way that we're onboarding. So by by asking for access to those systems, we can start to build the map, but it's an evolving process. So it's asking for information up front, but that's really the start of the process, because there's going to be questions that you or others would ask. in that onboarding process or in using the data that'll come back as, hey, we don't have that yet. And so if it comes back and tells us, hey, they don't, that information isn't included in the repo, we'll come into the next meeting saying, hey, can we connect to this random piece of software that you've got on the shop floor? Because that would really help us understand your question over here. Okay, one minute left. Yeah, one more question. Yeah. So I think we're talking about a lot of data here, and even in the organization, there are some data which is I can work on something like this sensitive. So how do you ensure there's data security and that the data is not going to the wrong and going outside? Yes, a very good question. So one of the things that, and I'll go back to the N8N example. The way that we're actually pulling information down from your business, the reason that we've chosen to use N8N instead of make.com or Zapier or other let's say, cloud-hosted applications, is that N8N can be put on a local machine and spun up on a local server and be fed this information, or you can query that information without leaving your corporate network. So by, you know, if it is, let's say, confidential information that you don't want to leave the premise, we put an N8N installation on site. on your network and we lock that down so that it's really only accessible from a certain set of computers or things on your network. or maybe using some other AI tools. Most of the time people are having concern that maybe this data is going outside. So like using GitHub, you have that capability that Yeah, GitLab is another option. So if GitHub is, you know, if you don't like Microsoft and their GitHub options, GitLab can be hosted locally on that same many-to-end server, and then you would just have your repo stored there locally instead of... actually on the cloud. With that, we'll say thank you for your time and listening today, and wish you the best.