 is caffeine connect. How many of you have received a call and realized about a minute into it that you are in fact talking with a robot? Artificial intelligence is a part of most people's everyday lives, whether it's an algorithm that determines the content you see on your social networks, the music that plays on your app or the emergency application of your cars breaks when you're about ready to rear in someone while you're checking your social networks and listening to music. AI is everywhere. Our next guest is here to help entrepreneurs understand how AI can be used in ways that make business owners more effective. By helping to demystify AI's role in companies and our lives, we can discover uses for artificial intelligence that will elevate our operations. Sean Meday is a veteran of the United States Air Force, a current intelligence officer with the Michigan Air National Guard and an engineering manager with a little company you might have heard of called Google. He's also a very good friend of DAV and his fellow veterans. In 2015, Sean was the first full-time team member for Patriot Boot Camp and those efforts have led to hundreds of veterans pursuing their dreams of founding companies. He continues to help DAV Patriot Boot Camp today by volunteering his time and consulting with us as we look at ways we can elevate the services we offer founders. So please feel free to post questions as we go and we'll do our best to get to them after Sean's presentation. So without further ado, to talk about artificial intelligence and layman's terms, I'll turn it over to Sean Meday. Thanks, Dan. I really appreciate the opportunity to be here and then to reconnect with so many great members of the Patriot Boot Camp family. I'm gonna go ahead and share my screen here. So if you can give me a verbal confirmation when you end up seeing it there, Dan, it should be coming across any second. We got it. All right, perfect. The technology hasn't failed us yet. Well, great. In our time together today, I'm hoping to kind of give you an explanation and background on what machine learning and artificial intelligence really are. My goal there is to really kind of demystify some of the hype versus the reality. And then talking about reality, let's look at some actionable use cases, some ways that entrepreneurs can leverage this technology to either supercharge their current venture or maybe inspire some future entrepreneurial pursuit that ML makes possible. And then we'll finally just give you some quick thoughts on what it takes to get started and where you might wanna go. And so with that, we'll dive into the content. Why is machine learning a topic worth discussing? Well, I think it's hard to go a day without hearing something about algorithms or the use cases that Dan highlighted in the intro. And that's not just kind of qualitative observation that maybe we're hearing more about these terms but we can actually go to Google Trends and see that quantitatively interest in the term machine learning has increased significantly over the last 10 years. So a lot of people are asking questions about what this technology is and what it means. I really like this definition for machine learning. It states that the ability for computers to learn without being explicitly programmed. The reason I like this definition, there's lots of definitions out there is because this one has really stood the test of time. This guy, Arthur Samuel kind of coined this explanation or this definition of machine learning way back in 1959. And I think what this highlights is the reality that for a long time people have been thinking about ways that technology can make their lives easier. We think about the Jetsons, which aired in 1962. The Jetsons had this amazing character Rosie the Robot and Rosie the Robot would take the kids to school. She would clean the house. She would do the shopping. All of the minutiae in our lives that we have for a long time envisioned that technology could someday alleviate. The reality though, I think is we don't have Rosie the Robot yet. At my house, we have a couple Roombas and those silly things get stuck on a daily basis. We're not quite at this reality of Rosie the Robot. So if we're not at Rosie the Robot, where are we? Well, I like this definition from a few years ago from a guy named Andrew Nang. And what Andrew kind of presents here is this idea that if a typical person can do a mental task with less than a second of thought, we're probably at a good place right now to automate that using AI, right? And so those are the kind of bookends I like to create here, right? We're not at Rosie the Robot yet. Hopefully someday we'll get there. Where we are today is really automating some of these split second decisions that cause a lot of perhaps human toil or rote memorization. You'll hear a lot of different terms in this ecosystem, right? We've talked about artificial intelligence and machine learning. You'll hear people mention deep learning, neural networks, natural language processing. It can all get a little bit confusing. And I like this comic from an organization called Signify there. You see their logo in the lower right. But I like this comic because it helps kind of create the hierarchy in my mind, right? Which is that artificial intelligence is the science. It's the umbrella. And machine learning is one example of the algorithms inherent in this science, right? So you may hear me kind of interchange the terms today. Just know that generally we're talking about AI as the umbrella and machine learning as the set of algorithms that is enabling the capabilities we're gonna talk about. Before anybody jumps into AI and ML though, the reality is they have to kind of already have some discipline or some practice around analytics, right? And I love this quote from Harvard Business Review because it kind of states that if your company is not good at analytics today you're probably not ready for AI, right? And I think this just kind of sets down the expectation that you have to have a coherent data strategy. You have to be collecting data. You have to be storing data, you know in a uniform way in order to really start to make use of machine learning capabilities. And I like to broaden this to really say organization, right? Because I think so much of the power of AI and ML comes from nonprofit organizations. It comes from government organizations, right? And so we really need to writ large kind of promote more data literacy in order to really start to unlock some of the power of ML. And the reason we need to do that is because the world is awash in data. We're generating more data today than we did yesterday and that trend is gonna continue going forward, right? We're expected to have 175 Zettabytes of data by 2025. Zettabyte is an incredibly big number. It's amazing to think that, you know from our humble beginnings, you know less than 100 years ago, we're now generating so much data. And the reason data is critical to AI is because it's truly the basis for how these machine learning algorithms learn, right? And so we've got to get proficient in understanding data in order to be successful in this ecosystem. We often talk about ML being the rocket engine, right? It's what's gonna take us to that next frontier in our entrepreneurial endeavors or in our technology pursuits. And if that's the analogy we're gonna use, data is definitely the fuel, right? But I'll caution you, just because an organization has piles of data doesn't mean they're necessarily ready to jump in and tackle the challenges of ML, right? Just because you have a shared drive on your organizational network that has lots of data doesn't mean you're necessarily ready to dive in and start leveraging artificial intelligence. But again, I'll just leave you with the thought here on this topic that data really is the building blocks of insights, but you have to have that data organized, you have to understand where it's flowing from and you have to have some literacy in terms of being able to slice it and dice it. If you take nothing else away from this presentation, I would just challenge you to think about machine learning as nothing more than pattern recognition. That's it. Machine learning is pattern recognition, but it's not the patterns that you and I as humans can quickly discern and quickly process cognitively. It's actually much more subtle and latent patterns, right? It's multivariate patterns that live in data, things that are perhaps too complex for a human mind to necessarily pick up. But that's really where this technology shines. So in the past, if we were gonna try to categorize different classes of data, we would use a rules-based approach. And what I mean by that is if we were given the task of categorizing the two objects you see on your screen there today, we might write some rules, right? We might say, all right, well, if the object is red, it's an apple. If the object is orange, it's an orange, right? So we've created two rules, our rules-based approach to classifying this data. The challenge is we can almost always find some input data that's gonna break those rules, right? So now all of a sudden, if our input comes in as black and white, we need to write another set of rules. So now if the object perhaps has a smooth surface, it is an apple. If it has a porous surface, it's an orange, right? We've now doubled the complexity of our rules. And, but again, we can find some input data that's gonna break that rule set, right? So here's a mango and all of a sudden our rules are broken. That's just kind of a recognition the real world is messy. Data doesn't always fit perfectly into categories. And I love this example, right? Because it should be easy to distinguish a dog from a mop. But even us as humans can have trouble doing that, right? We can introduce a sheepdog here and the seemingly easy task of categorizing a dog versus a mop gets a little bit more tricky. So that's where machine learning comes in. We're not gonna get technical today, but machine learning is all about matrix multiplication. It's all about breaking down data sets and looking for those pattern series and trends that will allow the computer to start making predictions. It's not all that different from how we as humans learn, right? I have three young daughters at a young age. I could ask any of them, hey, does this animal fly? And they would say, yes, this animal flies. Why? Because it has wings, right? And I think this represents kind of the first pillar and how we learn, right? What we do is we learn generalizations, right? We identify generalizations. If something has wings, it flies. The other part of that though is that we memorize exceptions because here's a winged animal and we know of course that this little penguin can't fly, right? So we learn these generalizations. We memorize the individual characteristics of the deviations and that's how our minds work and that's really how this machine learning technology works, right? And so you'll hear people talk about a neural network. All that means is that there is a series of technology gates that are breaking down data and looking for those patterns and ultimately coming up with a prediction or an inference, right? That's as technical as we're gonna get today. But what I wanna do now is kind of talk about how what you just heard behind the scenes can be actioned into interesting use cases, you know, similar to the ones that Dan mentioned, right? Things that are impacting our daily lives on a regular basis, whether we know it or not. I'm gonna break these down into four categories. We're gonna talk about natural language processing, we're gonna talk about speech to text, we're gonna talk about computer vision and then kind of this catch all category for like pattern recognition and anomaly detection and we'll jump into them now. We'll start with this natural language processing. What is it? Well, this is simply the ability for an algorithm to break down long or short form text, right? So you see a sentence in front of you, two sentences in front of you. What natural language processing can do is it can start to identify the people, places and things inside of this written word, right? So now we've got some categorizations, we've got an understanding of what composes this message. Behind the scenes, it's also doing a bunch of dependency parsing, right? So which words, which verbs, you know, are dependent on one another in a sentence. All of this starts to create some really interesting inference into how language works, right? For instance, when we do this on a large scale, we can start to see that the way the word man and woman are used in our written language are quite similar, honestly, to how the word king and queen are used, right? So behind the scenes, this matrix multiplication, these advanced algorithms are starting to learn about the relationship between distant objects based on the context they're used in. What does this look like in practice, okay? One really cool feature that Google rolled out a couple of years ago is called smart reply. This is a feature in Gmail used in the mobile version of Gmail. Somebody sends you a message and you're gonna reply. Smart reply will give you three quick replies that you may wanna send. Now, obviously if my mom sends me an email and expects a more personal response, this probably isn't the use case there. But if someone's just asking me to confirm an appointment or verify that some information is correct, the natural language processing here is able to kind of suggest these three short replies that might make sense. And what's cool here is to date, 15% of all replies sent in Gmail are using this feature. So you think about the user base of a product like Gmail, billions of users out there, right? If you can save a few users a little bit of time, the aggregate cost savings for an organization or for the entire population is pretty significant. And I think that's a great demonstration of the power of ML, right? If we can use some of this predictive capability to remove a little toil in the lives of our users or our customers, there's tremendous opportunity there. Another thing natural language processing does really well as sentiment analysis, right? So these algorithms have gotten pretty sharp at understanding based on what's written, based on what's there in text, is the sentiment positive or negative? And what is the magnitude of that sentiment, right? Is am I really mad or am I kind of mad? Am I really happy or am I just kind of moderately happy? So we can start to do some empirical scoring of text, right? A really cool use case for this is if you're getting a lot of inbound email from customers, being able to start empirically scoring what's positive versus what's negative. And that leads you to secondary questions. Like, well, what topics in these positive emails we're being referred to? And you can start to create an interesting knowledge graph of what you're hearing from your customers, what you're hearing from your partners and segment that data to start to really understand a landscape empirically in ways that we couldn't just a few years ago. Chatbots are another byproduct of this technology. And I imagine you've seen them, banks use this a lot, but this is a way for a website, an app to try to derive intent from what a human types and try to point them to authoritative data. Cool thing about Chatbots is you can really start to integrate some of the language translation capabilities that have become so mature inside of machine learning. And so you can appeal to a wider audience, right? If you know you have a constituency that's multilingual, this is a great way to try to meet these non-English speakers where they are and try to provide more relevance for your platform. All right, we'll pivot here and go talk a little bit about speech to text. Speech to text is, again, the ability for a machine to hear spoken word and to transcribe it at its most basic sense, that's what it is, right? This has really become popularized with these in-home devices like the Google Home you see here, certainly Amazon's Alexa. What's really staggering today is there's a stat that just kind of, I think blows my mind on this, right? Which is that 20% of all Google searches coming in today are coming in via voice. So we can think that an entire generation of users is getting accustomed to having conversational interfaces with technology. And today we primarily got this in our homes, maybe our phones, but it's interesting to think about how that expectation of having a conversational interface into technology could permeate into our professional lives, right? So sitting at the desk there, this is the ability to ask, what is my current PTO balance? Hey, I'm going to Atlanta next week. What's the per diem rate there in Atlanta? To date, these are questions that come up in office spaces all over the world and generally it requires a human to go log into a system or to search a system and pull back authoritative information. The value proposition for this speech to text technology is that it can start to take some of that toil away and not require the human to put hands on keyboards or think about necessarily where they're going to search for this information. So it's just tremendous opportunity here for entrepreneurs to take this maturing technology and start to apply it to interesting use cases that their customers might be struggling with. I had a really cool opportunity a couple of years ago to start working on a side project. Google has this concept of a 20% project, which is you get to spend some of your time working on something you're passionate about. For me, this was about trying to take some of the technology that I saw in the ecosystem and apply it to an interesting problem that was close to my heart. And that's, in this case, what manifested was a really neat technology, a really neat set of capabilities that allowed me to transcribe public meetings and hearings. And so I live in this little town of Superior, Colorado, and we're only about four square miles and 13,000 people. But our town government has 300 hours of meetings and hearings every single year. And I care about my community, I care about my little town, but I can't go to 300 hours of meetings, right? So what I was able to do here was take video from these meetings, run it through this speech to text model, and come back with a full transcript. Now I had a searchable transcript where I could start to say, all right, well, I'm really interested in the new signage ordinance, or I'm interested in what we're gonna do about standing up a new community center. And I could then search this hundreds of hours of videos for just those seconds that were relevant to what I was looking for. So really cool use of this technology, I had a lot of fun, and then Google let me open source all the code. And some other folks have taken it and run with this project, so I'm pretty proud of this one. All right, we're gonna pivot now to what I would argue is perhaps the sexiest of the machine learning, kind of opportunities that we're gonna talk about today. And the reason I think this is such an appealing one is because computer vision is really about pixels, right? And we are visual creatures, humans are visual creatures, and computer vision is about making sense of all these pixels. My first exposure to computer vision came almost a decade ago when I had a chance to work on the Google Earth and Google Maps team. And at the time, Google was spending a lot of money and engineering development research into the Street View program. Street View was really about trying to get out and drive the streets and collect high-resolution imagery so that somebody who was gonna embark on a journey from home to a shop or home to a trip could start to understand what it looks like at their destination before they make the investment of time or put shoes on the ground, right? What was neat though is that once these computer vision models matured, Google could start doing really neat analysis of all these pixels, right? So you've got high-resolution imagery of a street corner as you see here. But then this machine learning capability was able to start pulling meaning out of those pixels, right? Could it identify street names? Could it identify traffic signals? Could it identify business logos, right? So a neat way to take a huge corpus of pixels and start to make it actionable, start to make it searchable, start to make it understandable in different ways. Another really cool project I got to see come to fruition at Google, didn't work on directly, but was super exciting to see was a project they called Pretty Earth. And if you think about satellite imagery, collecting satellite imagery, there's almost always a day where you're gonna have cloud cover or atmospheric haze somewhere on the earth, right? And so it's really hard to get a cloud-free mosaic or a cloud-free picture like you see here. This just doesn't happen in the real world, particularly in the Southern Hemisphere and the rainforest of South America. It's very rare to get a clear shot. So to make this clear shot, what Google did is they worked with the US Geological Survey and NASA to go back and pull 60 years worth of satellite imagery and then find the one perfect pixel in each of those 60 years where there was no cloud cover and then mosaic it or combine it all together. The other reason I really like this project is it demonstrates some of the scale, the crazy scale required to pull off these grandiose machine learning projects. As you sit here today watching on your phone or maybe on your laptop, that device probably has one, maybe two CPUs, one or two processors that are making all this computation on your device happened. To make the Pretty Earth mosaic you saw in the last slide, it required 500,000 CPU hours, right? 500,000 CPUs working together in concert for one hour or some combination thereof, right? 100,000 for five hours, whatever the math was, that's a lot of compute power. And I think that is a really cool example of what machine learning at scale starts to look like. These features have permeated into our daily lives as well. I'm a big user of Google Photos, I have a young family. As we travel around, I snap a lot of photos and Google Photos uses this technology to make my photos searchable, right? So if I want to think about that time we were at the water, I don't have to go through and tag all my photos with water. The machine learning models have matured to a point where they can recognize several thousand different people, places and things inherent in these pixels. So a really cool way to augment the user experience, take a bunch of pixels, make it searchable, make it discoverable. My team at Google got to work on a cool project last year with the Navy, which was using some of this computer vision technology to look for, identify and classify corrosion and rust on naval vessels. The Navy spends about a billion dollars a year fighting rust and corrosion on their fleet, right? And this was a cool way to use overhead imagery, right? Fly over a vessel while it's in port with a drone or a quadcopter, take that imagery and start to identify where the rust is and how severe the right rust might be. In the past, this was a very manually intensive process. You had to send people out to go walk the ship or boat around the ship and visually inspect. What this allowed someone to do was come through, train the imagery, right? Go through some images and say, yes, this is rust, this is bad rust, this is good rust, then turn that training data over to an algorithm and let the algorithm start finding those patterns, series and trends in the pixels that identified rust. So really cool use case and certainly near and dear to the heart of our naval brethren and sisters here on the call today. All right, and then they might catch all category I warned you about before, pattern recognition and anomaly detection. This is really about finding those latent patterns inside of data. And so this provides really cool unique capabilities for doing things like smart scheduling or equipment loadout, right? If we're gonna send a team to the desert, we're probably gonna send them with different equipment than if we're sending them to the Arctic, right? And so if you can start to identify what different configurations might look like in a data sense, you can start to do really cool matching. You see this today with platforms like LinkedIn, right? Where they're looking at your talents, your interests and they're trying to match you with jobs, they're trying to match you with other professionals. What's going on behind the scenes there is a machine learning algorithm that's just trying to make sense of the signals and the input that it has received. The other side of this is anomaly detection. This becomes super powerful, right? If you've ever received one of those alerts from your bank saying, hey, we saw this charge in your credit card, it looks a little suspicious. Did you make this charge, right? That is generally a machine learning model that behind the scenes looks at your spending behavior, looks at your demographics, looks at the time of day and starts to identify anomalies to your common pattern of life. I had a call a year or so ago from the bank asking if I was trying to spend $4,000 at a dollar general in Tennessee and I was not. And I do have to wonder how many carts of goods you have to fill at a dollar general to spend $4,000. But that aside, behind the scenes, what was happening was a machine learning model was looking at all of my spending history, perhaps my travel, and it identified this was an anomaly. Tennessee was an anomaly, the amount was anomaly, perhaps the vendor that I was trying to spend with weren't an anomaly. But as entrepreneurs, there's so much opportunity right now to apply this type of machine learning model to interesting use cases around things like fleet management and predictive maintenance, right? Sensors are relatively cheap and so as manufacturers are building equipment, they're loading them up with sensors, inside of all that sensor data that's being collected is an opportunity to start to identify deviations, right? When, what does it look like when a healthy set of brakes is operating versus what does it look like when a set of brakes that needs to be replaced is operating, right? Somewhere in the data collected there is a signal about the opportunity to get ahead of a maintenance failure. And then you see this a lot in the IT space, right? So as we think about identifying cyber attacks, as we think about identifying malware, ransomware and exfiltration of data, generally those attacks are detected through the analysis of log files, right? The logs that are behind the scenes being created every time you and I log into a website, log onto the internet, open our device, close our device, turn our device on, all of those signals when looked at in combination can really demonstrate interesting patterns and then allow us to find deviations from those patterns. One really cool project that the National Oceanographic and Atmospheric Administration worked on with Google a couple of years ago was looking at underwater audio that had been collected over 20, 21, well, it had been collected over 14 years but it was multiple microphones. And so in aggregate, this represented 21 years worth of underwater audio, right? If one person were to sit here and hit play on all this audio, it would take 21 years to hear it all. But what the researchers were able to do was create a machine learning algorithm that could listen to all this audio and identify where it heard humpback whales, right? And again, so listening to the standard noise of the sea and then finding deviations from that noise and then differentiating those deviations from other noise sources, right? Is it a ship? Is it weather? Or is it truly a whale? And this was really neat, very successful, ended up 98% success rate identifying humpback whales inside of this large corpus of data. Another area you see this capability applied is in data centers, right? And you can think about almost any industrial scale operation. If you can really instrument your equipment, you can instrument your electrical use, you can instrument your HVAC use, your heating and cooling, you can really start to identify patterns and then you can, from those patterns, find ways to optimize, right? Find ways to save a little bit of power. What Google has done with this and with the data centers is really look at what is the ambient temperature outside? What is the expected usage? And then how can we tune the HVAC systems and the power systems accordingly to try to save money that we can also pass on to our customers and shareholders, right? So I think as you are looking for new and interesting ways to optimize things for your customers, this capability provides some really promising results. All right, so those are the four big categories we wanted to go through. The final thing I just wanna do is leave you with a couple of thoughts about where to get started. There's lots of companies out there doing interesting things with AI and ML. You have some interesting opportunities to try to bake this into your existing products or create brand new products based on this capability. And so as you think about what it's gonna take, there's some different avenues you can pursue here, right? And it's all gonna be based on kind of the composition of your team and how much flexibility you're looking for. And what I mean by that, and let's dive into that a little bit, right? If you're just gonna do things like identify dogs versus cats and imagery, a lot of that work has already been done and that's kind of a solved problem. And you can use some pre-trained ML models, right? Kind of low development time. You can start incorporating that feature into your app or into your system very soon. The speech to text example I gave you with the public meetings and hearings is a great example of this. That was one where I didn't have to go train a new machine learning model. Companies out there have speech to text models that'll do transcription. I just took that model and combined it with a new use case, right? My use case was the meetings and hearings. And that really, that's the innovation available here is recombining some of these existing capabilities with new and exciting opportunities. But those pre-trained ML models aren't particularly flexible, right? So if you really need to do something custom, if you're trying to do something nobody else has done before, that's what's gonna require the custom ML model development, which is, we'll talk about a little bit more on the other slide, but generally requires a lot more development expertise, more development time, but ultimately gives you a lot more flexibility. So you have to kind of figure out where you are on the spectrum, look at what's out there. There's a cool class of tools called AutoML that Microsoft, Google and others have brought to the market, which is really about trying to meet the developers where they are, right? You don't have to be a machine learning expert, but if you come to the table with some training data, can we help you start to create a accrued machine learning model from that training data? So as an example, I played with an AutoML demo a few years ago where I just took pictures of fighter aircraft. I took a hundred pictures of fighter aircraft and I wanted to see if I could train a machine learning model to differentiate an F-16 from an F-18, right? Something easy for a human to do, we just look and see does it have one tail fin or two tail fins, but I wanna see if a machine learning model could do that and AutoML gave me a way to do that. I just found a hundred pictures of images, I classified them, I said this group is F-16s, this group is F-18s. I uploaded them, hit a button and AutoML kind of created a model, right? So it wasn't quite as off the shelf as that dog or cat model we kind of, talked about hypothetically a minute ago, but it was much easier than having to go out and develop our own custom machine learning model. And if you are gonna go there out of creating a custom machine learning model, right? Maybe you have the new cool exercise app and you're gonna collect interesting data about exercise. Well, you're probably not gonna find a pre-trained model or an AutoML model that's gonna fit your need. What you have to be thinking about is you're gonna need a large training data set. You need to train the algorithm in order for it to make the predictions and the inference about what's happening. And so you gotta either collect or start to generate a large data set. You know, the models take tuning, these things don't work out of the box, right? There's a little bit of complexity involved with trying to figure out how to get a consistent result from your inference and prediction based on a machine learning model. And then as we talked about with the pretty Earth example, this can take a lot of computation, right? And so unfortunately sometimes that can get a little expensive, but that's a conversation for another day. So that's kind of what I wanted to share. My goal was really just to get you thinking about the art of the possible. Demystify some of the hype, separate some of the science fiction from reality and then talk you through some real world use cases for this technology. And I appreciate your time and would love to kind of, you know, open it up to any questions that you think the audience might have, Dan. That's great. Sean, we have someone joining us on the Patriot Bootcamp site and alumni, John Denning. And he asked, when an app like Google Maps forecasts an ETA for a trip that he's taking, he said, what categories of machine learning are you using? And he said that it seems to be getting a lot more accurate lately. Yeah, so that's a great example of, you know, kind of the anomaly detection and the pattern matching we talked about earlier, right? Is looking at what does it normally take to drive from point A to point B? And then what does it normally take to drive from point A to point B at a certain time of day? And, you know, having spent a couple of years on the Google Earth and Google Maps team, I'll share one interesting story with you, which is initially, when that feature was rolled out, where we'd say, hey, to get from your point A to your point B that you just told us about, the fastest way is to go, you know, whatever, up the highway, we found a lot, they found a lot of skepticism from the users. The users would go, come on, this thing doesn't know better than me. And so what the team did, they did a bunch of usability testing and they identified that if they showed three routes total, but the one that they thought was fastest, it would de-risk some of this in the user's mind. The user would see the three routes and go, oh, well, I think the algorithm has talked this through. And it made the users more accepting of that suggested route. So, you know, I like that example because it reflects that this isn't just a technology problem, right? This is a user experience, user psychology challenge on top of providing good technology. So great question. And, you know, there's, I would just call that a multivariate problem. It's taking into account a lot of different variables to try to give you an accurate estimation. Thanks, Sean. Google for startups is a longtime supporter of veteran entrepreneurs in Patriot Bootcamp, which is great. One of the questions John had as an alumni was with Google Cloud Platform. And he wondered, is there a resource or individual that you could connect folks with that could help them evaluate that versus a competitor's services? Yeah, the Google for startups program is definitely the right point of contact there. They can provide, you know, credits and resources to get you started. That's all kind of part of their charter and mandate. If you're not getting what you need from them, you know, I'm always happy to try to point somebody in the right direction. So, I will make sure my info is available. And if you're not getting what you need, hit me up on Twitter or LinkedIn and I'll try my best to point you in the right direction. But yeah, generally the front door for that sort of stuff is the Google for startups team that Dan mentioned. Yeah, he was actually asking though, I was just thanking Google for startups because it's great that we can connect people with funding. But he's actually asking about the Google platform. You know, he's got people sometimes get married to a certain kind of technology and his team likes another technology and he's just trying to see what he can do to kind of turn them on to that and wondered if you had any advice on that. Yeah, I mean, you know, for a long time, I think as technologists we fell into the trap of like a winner take all kind of mindset, right? Where, you know, and there's also some law of primacy. What you learn first, you learn best, right? So, if I start out on platform A, I've got a lot of some costs there, it's what I know. But increasingly we live in a multi-cloud world, right? And I always like to point people to our personal lives, right? I mean, I sit here with an Android phone, my wife has an iPhone, my kids have Amazon Kindle fires. We have a Netflix subscription, a Hulu subscription, like we live in a multi-cloud world, right? We buy services, you know, to get the best of each vendor. We buy from different services. And I think, you know, cloud is the same way, right? If you're just buying from one cloud vendor, you're probably not getting optimized services across the board. And so, you know, that's where we really start to think through a multi-cloud strategy. How can you get your folks kicking the tires on a couple of different cloud vendors to see which of these models, you know, and here I'm being specific to machine learning, but which machine learning models or those pre-trained models might work best for your use case. And so I think it's specific to what you're trying to solve and then really looking at who the vendors are in that landscape and then finding a way to quickly evaluate which of them is going to be the most performant and the most cost-effective for your need. That's great. You know, a lot of founders, they're starting off, they have small budgets. And one of the questions we had here was, do you have any tips on how they can apply machine learning, a smaller company, to make them more efficient? And then it's kind of a two-part question. And is there a specific application you recommend for companies with small budgets? Yeah, it's a great question. And you know, it's also reality in the world we live in today that folks who really know machine learning well, data scientists, machine learning engineers, they are in high demand right now. And so generally, you know, loss, supply, and demand is that makes them very expensive to employ. And so, you know, there's a suite of tools out there that are really focused on trying to meet the users where they are, right? So if you can't go out and spend $600,000 a year to hire a machine learning engineer, you know, what can we do to take your current data analysts and make them more effective, right? And so there's a lot of companies out there that are trying to roll some of these machine learning features into their existing tools and platforms. And of course, you know, my employer is Google, I'm a lot of Google, and I also just know Google products better, right? So I'm not here to pitch Google, but a couple of examples I'll give you, you know, we have a data warehouse called BigQuery where you can just dump lots of data. Well, they've rolled some machine learning models into that. And so with the click of a button, you can start running machine learning data, you know, models on the data that you've uploaded. I think that's a good example of an existing vendor rolling in. We have another platform called Vertex AI, which allows you to do some of that more custom modeling we talked about. But I would just survey the landscape, see what's out there, you know, see what tools you're currently using for data ingest and data storage. You're just doing it Microsoft Excel today. You know, there aren't a lot of tools there, right? So maybe you have to think about using a different platform, a Tableau or something like that. But, you know, there's lots of startups out there who are identifying this as an opportunity, right? How do we take these great machine learning capabilities and meet the entrepreneurs, meet the users where they are today so that we can reduce some of the complexity involved with getting started here? You brought up the Jetsons earlier, which is my speed. That's about my intelligence level here. But what do you say to founders who are concerned that they're gonna use AI, especially in customer service kind of areas, and an outward facing application is gonna make them look like they just don't care or hurt their reputation in some way? Yeah, that's a great question. And I think, you know, it gets to another kind of critique or concern I hear sometimes, which is, you know, is AI gonna replace employees, is it gonna replace humans? And I think, you know, what I really wanted to try to communicate with that comparison arose in the robot is we're nowhere near there today, right? We have to recognize this machine learning technology isn't what we want it to be. And instead of replacing some of our humans, I think what we need to think about is how we can augment our current human operators, right? So, you know, if you have one support person and they sit there and they respond to support requests from your customers or your users, you know, we're not gonna replace that person, but can we increase their aperture, right? Can we use some of that natural language processing to help them derive some intent from our support request before they even open it? And start to make some smart connections between that support request and other support requests, right? So that, you know, if that human could read seven support tickets a day, you know, previously, can we get them to a point where they can triage 100 support requests a day, right? And I think that's the power of AIML is not to replace our humans and not to, you know, try to provide a human-like experience. Instead, let's increase the aperture of our human analysts, let's make them more effective so that they can, you know, handle those customer service type scenarios that, you know, really only a human is equipped to handle. On that note, you know, a lot of people are scared. They're nervous. They're seeing the science fiction side of this and they're scared about the amount of their information that's out there. And in some cases, that's going to limit possibly their capabilities as founders. What do you say to veterans who shun technology because they think that it's either invasive to their customers or that it's a surveillance scam from the government or something like that? Yeah, I mean, there's always, you know, those concerns and there's definitely legitimacy, you know, concern or legitimate concerns about, you know, the privacy of individuals. And so I think what it comes down to is you really have to know what data you're giving a platform and then you have to really understand what their terms of service are, right? And I think, you know, we all tend to click through pretty quick whenever we join a new service and it says, you know, here in terms of service, do you accept or decline? And by the very nature of the fact that I just want to use the service, I'm inclined to click accept. But as you're making those decisions for your company, you know, you really want to understand what you're agreeing to and where that data is going to ultimately reside, right? It is, if you're going to store your customer data in somebody else's platform, you want to have assurances that they're not going to use that data for any purpose that, you know, you might deem abusive or hazardous and various, right? And so I think it's really about, you know, looking at the topology, where's your data going? Who's touching your data? And who in terms of what vendors? And then what are the guarantees those vendors make? And sometimes that requires, unfortunately, diving into some of those legal agreements or, you know, contacting your friendly neighborhood lawyer who may be able to translate that legal ease a little better. Kind of on that note, probably with user terms, Stephen asked, he said, Google's strict with apps that listen to phone calls. Do you have any tips, he asked, for navigating those policies? No, you know, I don't have a good answer to that one. That's a little bit outside of my field of expertise and understanding, you know, I will just say that Google is one company that spends a lot of time really trying to make sure that people aren't going to use the platform and make sure that people are going to be good stewards of user data and customer data. So I'm sorry, I can't help more on that one. That's fine. Sean, do you have anything else before we close up here? No, just I really appreciate this opportunity. And you know, as I think back over my professional career, the time I got to spend with the Patriot Bootcamp community is really just an experience that stands out. And, you know, the important work these entrepreneurs are taking on, the changes they want to make in the world are just inspiring. And I'm really excited to see Patriot Bootcamp, you know, under the umbrella of DAV and the great things this organization is going to accomplish in the future. You've got a great team, great alumni base and I'm humbled to be a small part of this amazing effort. Well, you're a huge part of the effort and we're extremely grateful for you. And it is what we're learning about Patriot Bootcamp is it's definitely a community and it's a great community because of people like you and the willingness that people have to, well, they use the program and then they take the program and they help other people out and they go from being alumni to being mentors and then they come back if they need a little bit more help with something and go from there. Everyone, very grateful for your participation, Sean. So thank you and hopefully we'll talk again real soon. And thank you to everyone for tuning in. Our next caffeine connect, John Denning who passed the touring test, a fascinating person is going to be talking about app development and technology use in companies. You can sign up for updates on DAV's entrepreneurship efforts by signing up for a newsletter at patriotbootcamp.org. And if you have any questions, Sean is sincere when he says he's willing to help people. If you have any questions, please email us at info at patriotbootcamp.org and we'll go ahead and forward those questions along to Sean. Thank you so much for your time. Thank you to Google for startups for the great support that you offer Patriot Bootcamp alumni and...