Transcript 

Brandon Carter:

Good morning and good afternoon, everybody. Thank you for tuning in. Welcome to our webinar.

We are jumping in on session two of our AI webinar series. The first session really focused in on giving your employees access to AI tools and how to go about that.

But today we're going to be focusing in on how does AI help you with your products? How can you integrate AI in your tools, in your services, in your platforms?

Just a quick overview of the agenda. We have a handful of specialists and experts that are going to be discussing the topic and throughout, we're going to take your questions. So normally we would reserve Q&A for the end, but for those of you that are here live, we'd love to get you to interact, to hear from you. We want to hear your questions in real time. And we'll address most of them in real time or get to as many of them as we can.

My name is Brandon. I am the Marketing Director at Promevo. I'll be your host and moderator today.

But, originally we were going to have a guest from Google join us. Due to unforeseen circumstances, he's not able to join. We're going to produce that conversation in an on demand environment here in a couple of weeks. So keep an eye out for that.

But instead we've got a handful of Promevo experts who have been in the trenches with our clients and with companies throughout their careers, building out AI tools, building out software, and AI related development.

I'm going to go ahead and have them introduce themselves. And let's start with John.

John Pettit:

Hi, everyone. John Pettit, CTO here at Promevo. I've been in the software development space for 25 years, and I've really seen the development of AI from, the early beginnings to where we are today, which is pretty sophisticated.

I'm happy to talk about different ways that this might be able to help you today.

Justin Barone:

I'm Justin Barone, Principal Cloud Solutions Architect here at Promevo. I've been working in the cloud since before it was called the cloud. Over the past couple of years, I've been implementing a lot of different little AI projects for our clients.

Real interesting stuff. I find it fascinating. I'm excited to see where AI is going to take us into the future. And, I'm excited to see where this discussion goes and talk about some of the frequently asked questions people have around AI. So excited about this.

Aaron Gutierrez:

Hey everybody. My name is Aaron Gutierrez. I run the data engineering and analytics projects for Promevo. Everything from BI solutions all the way to, and including AI projects

I'm also working with Justin under a couple of these AI projects internally, but at the same time, I also like to dabble with AI features and tools on the side and a lot of parts, a lot of quick moving technologies coming and going these days.

So it's tough to keep up, but it's at least interesting.

Brandon Carter:

That's right. And the interesting thing about is again, like these are experts, but you're probably going to see them on your projects. These aren't just our expert to sit there and tell everyone else what to do. Like, you're going to interact with these guys on a daily basis.

So who is Promevo? We are a premier Google sales, service, and build partner. That means pretty much everything under the Google Cloud umbrella. We service your Chrome devices, Google Workspace, Gemini is obviously a big topic for us.

And we manage a platform called gPanel that gives you like a n automation and omniscience over your Google Workspace environment. So you can see security threats, you can see what your users are doing, you can control email signatures. If something happens in your Drive, you can get an alert.

Yeah, any interest in that, reach out to us. We'd love to chat with you.

And of course, we obviously do a lot of custom solutions as well. Google Cloud products and services are great, but a lot of times, It's not just taking them out of the box and setting them up. You need a little bit of help managing them.

You need support, and you need sometimes custom development, which is really where these experts that are here on the webinar with us today, really come into play. So reach out, promevo.com, we'd love to hear from you.

And just as a reminder, this is the second part in a four part series. A month ago we talked about just basically AI in your business, giving AI to employees, whether that is Gemini or whether that is like a BYOAI service where you're not really going to give it, but they can go and sign up for whatever service that they want.

Really interesting discussion. I think really eye opening and the kind of thing that every company to one extent or another is having to reckon with today. AI is in your organization today. That's the whole gist behind this, whether you're, in control of it, or on top of it, or not. It's here.

And what we want to do is help you understand it, understand the different scenarios. And today we're really going to dive into AI in your product. And that is AI not just your product, but what are your services? AI in, the way, like, the tools that your employees have? AI in your SaaS products.

With that, we're going to move into the part where we just talk and we discuss things and I want to get these fellows opinions on just from the first perspective, does every tool need an AI component? Is AI right for anything? If you have a service or a product or a platform, should you be considering AI today?

Justin Barone:

So my thoughts are, AI is good for very, for certain situations. Something that's going to solve problems, whether it's enhancing efficiency or user experience in your SaaS product.

Gemini, when you're using it in Workspace, right? Gemini is really integrated through all the Google products like Docs and Sheets. And those things are really great for helping you be more efficient and get your job done.

If you're building a SaaS product and you have your own tool. You're your own product that you're offering. You can start looking at ways, right? Of how can I put a I into this product?

That's, not just for the fact that, oh, is the hot thing right now, but actually, how can it, Solve those two things that I talked about, right? Enhancing efficiency. And then the user experience always comes down to making their life easier when they're using your product; maybe you're, you want them to be more conversational with pieces of your product, things like that.

So I always look at it from, not every tool needs AI, but you can probably find some little things where you could dabble in it, and proof of concept it out. That's my kind of thought on that is I think it can be, I don't think everything can, right?

Yeah, I'm curious what you guys think.

Aaron Gutierrez:

Brandon, you said a really key word to me, which should you consider AI? And I think answer is yes. Just because, if you're not going to at least consider it, your competitors are going to. So it's worth at least a heavy thought about the tools that are available for you.

And like Justin mentioned, some are just really easy to use out of the box. Like not everything is programming a RAG a LangChain from scratch.

Like, the Workspace stuff that's, the AI that's built into that super easy to use. If you want to start writing an email, it'll just " Clippy" you and it will say, Hey, it looks like you're writing an email. Like how can I help with that? So it's, especially if it's such a low barrier to entry, low effort, like there's no harm in, to check it out, read up about it, look for free trials from companies that offer that type of stuff.

But like I said, though, if you're not. If you're not at least considering it, your rivals are, and they'll use it to get a leg up in the business.

John Pettit:

Yeah. The way you phrased the question with every tool: hammers, I don't think we need AI, right? Yeah.

There's probably some areas where it doesn't make any sense, but everything that we use from an information software, world today is about productivity. And, the form that existed to make it more productive for people's get rid of paper and things like that. And so we have these digital forms that we fill out and people try and automate that with dropdowns and different forms of user experience to speed it up.

I think AI can speed up a lot more of that, right? People who are taking other forms of entering them into the computer, things like that could just be vision problems. People who could be speaking into the forms, to solve those things, workflows that could be using LLMs as like a natural language programming method for people who aren't programmers.

So, I think every kind of information management tool could probably use some form of AI. I think that the challenge for most people is going to be, what's the right problem to solve with the right AI tool at any given time? And then how do we do that effectively? Like how do you succeed when you make that first step?

I think we're just going to see it in other consumer software and business software, just de facto. And when we went to our last conference, everybody has AI in their software now. It's not, crazy new that people have been building out more automation tools and things like that. So I think we're just going to see a faster advancement of that.

And I think you're probably going to find new ways to make your life easier. We've all been looking for like an easy button. I think that AI could help with that if done correctly.

Justin Barone:

I'd also add to is the evolution of technology as tools, right? The hammer came after, probably a rock at, somewhere down the line, right?

We're not using abacuses anymore to solve math problems, that's another thing is the AI is there, right? And you can look for ways of utilizing it, but don't be, like, afraid of it from a standpoint of, Oh, you're going to lose this. When we got calculators, we didn't lose anything. We just got more efficient.

Brandon Carter:

Here in a minute, I do want to get into when you said afraid that, that's a really interesting topic and I think something that a lot of people are thinking about when we're talking about security and ethics, I want to table that for a second because it is important and we do need to talk about it.

I am curious though, like just for me, I want to talk about some of your favorite implementations. Like we've all got various tools where we see AI. Uh, for those of you watching this in the future, this webinar will be largely edited for on demand consumption by like an AI tool that will cut out a lot of my stutters and uhs and likes which is great.

One that I experienced the other day, Gemini has added the ability to interact with PDFs which is massive. We have dozens of PDFs in our Drive and to be, and some of them are huge and to be able to consume that information, condense it, summarize it. I'm just curious; those are two that I think are really they seem obvious, but yet we're just now getting those in the year 2024.

I'm curious if any of you have an AI implementation in a tool or a service that you use that like really sticks out to you.

Justin Barone:

The fact that you brought up being able to push a PDF in, and one of the things I love is that you can chat with your documents, right?

That's a weird concept to think about. You have all this data that you've entered in and you've got these documents, right? Or they're generated over time, whatever. But being able to input those and then chat with them, that's insane to me. And that's a really, that's really cool because I'm going to ask questions like, Hey, what was what, give me some statistic whatever's in the document. I can like ask and infer things and suss some stuff out of that.

That to me is really cool because that's where, I'm a big Star Trek fan. As you can see, I'm drinking out of my Spock cup. But being able to say computer and talk to the computer and tell it what that's really cool. That's amazing right there.

And, Aaron can probably attest to this. We're working on stuff where we're being able to talk with data sets, right? Have a conversation with your data. That's, and that might sound like that's crazy in the future. That's now. We're starting to do that stuff now.

Aaron Gutierrez:

Just to add on what Justin said, the crazy advances in LP processing of data and then feeding it a scheme of your database. Like those two things combined is what gives you the power to do that. And that's really cool to see how that's progressing and getting better as they keep working on the under the hood algorithms to, to make good solid queries.

I've seen terrible queries that it generates, but like at the same time, there's some great stuff that would take me 15 minutes to write a good query. It'll do it instantly for me. , that's really cool to see the progress there.

To answer your question about implementations. I just, I want to mention some of the quality of life stuff that I noticed just in various tools that already exist. I know me and Justin have been working on code, and one day I noticed on his IDE, he's got, like, his co pilot was recommending him what to type in the box after, after you write code, and you want to commit it to your repository, it's got the little message box that says what is this change about?

It had a feature to just do that for you. So it knows all the code changes you made and was able to write a little summary, rather than you sit there being like I changed this field. I added this table and stuff.

Like that's all to be clunky. But if AI can handle that little administrative part of things, like that's to me, a perfect implementation of quality of life enhancement that AI is like sneaking into all these little applications and products that exist. It's pretty cool to see those types of things weave their way into existing programs.

Justin Barone:

Yeah. And when it's doing that, it's using semantic commits. So not only is it looking at what your changes are, but it's also putting it in a format that's like a common format that engineers are using when they're committing to you know, source control.

Aaron Gutierrez:

I wouldn't do that. So yeah, I'd miss that. You've seen some of my commits.

John Pettit:

I really like the the multimodal side of things. So there's a couple of times where I've been able to just take a photo of something and have AI summarize it for me or grab some information out of it. I know there was one that, that we had a whiteboard session.

People had drawn a bunch of stuff on a whiteboard and took a snapshot. Tell me what this is all about. Without having to turn and interpret it or create notes on it.

Another one was. I had an app I was using and it didn't have the ability to write a report, but I wanted to get a list or a table that I could quickly put into another format, into a presentation.

I just took a screenshot of that and threw it into the AI and had it break out a list of project statuses and other stuff that you can get out easily some other way, or I would have to go into a reporting tool to run when it was already displayed on the screen. So, the multimodal side of it, I think is pretty impressive of how it can use, like, the vision information to interpret what's in a picture, not just text.

Justin Barone:

The have you, any of you use that Sheets feature where you can say, like, what kind of spreadsheet you're trying to do and you literally tell it that you hit up, hit the button, and it'll pop in some ideas with a kind of already flushed out and. Then you can just enter data into it.

That's wild to me. That's, that one was really cool.

Brandon Carter:

I'll drop in a pitch here that we actually published a video, I think it's on our YouTube channel, where I generated a marketing plan using Gemini, but then went through and revised it a few times. Like I need a marketing plan to announce our new gPanel functions. But then I went in and it produced, like, pretty well thought out extensive sheet. But then I said, all right, but my budget is like $10 K and we can only do online ads, completely revised it. It's really impressive. It, and that saved me hours. Of work. And like the thing with AI is it may not have been perfect right off the bat, but it saved me those hours of thinking through it gave me the ideas.

And as a result that much closer to getting a launched marketing campaign. That video's on our YouTube by the way, just a small pitch for the marketing content that we generate.

Aaron Gutierrez:

Stories like that are cool because you didn't, you didn't lose your job like it helped you move way forward than where you were. Like, I think when people hear those types of things, it'll make them less scared to use AI because it doesn't, it doesn't replace you. It enhances you.

Justin Barone:

Did you also ask things we built as well?

Brandon Carter:

Let's get into that. We can talk about, so the first part of this, I want to really give people the idea of, should you be considering this?

And what are some of the like rubrics that you need to consider? Should I look at implementing an AI component to my tool or this internal process that we've built for my employees? Should I have AI in it?

I think it's obvious the answer to that is yes. And, but it might require some creativity.

Let's talk about building, like what are some of the, let's start to move into that. That's the second part of it is what's it take? What do you do? So I don't know. I'd love to hear like from you guys, what are some of the, fun things that you've built are some of the things that you're most proud of.

Justin Barone:

I want to talk about some stuff I built cause I love talking about stuff I built cause I think it's, I geek out on this stuff.

We've done, we, like I mentioned before the, one of the things that really excited me was being able to chat with your documents and talk with your documents. So we've, we built some stuff around that.

The RAG solution. So retrieval augmented generation, being able to take generative AI, but also mixing that with API calls to systems and services, doing some really cool almost hybrid approaches to AI right there. That has been really amazing, right? Because then you're not just going off of what the model knows. You're feeding the model information, very specific information based on whatever you're, you're setting this up for and then you turn around and then ask it questions about that.

So it's almost very similar again about asking the questions. about your documents or your data. You can also, it's not necessarily just like it has to be a chat interface, right? It can also be programmatically.

You're filling out a document that has very specific questions that are being asked. Another thing we've done recently is a transcript analysis where we've taken a transcript of a conversation. It's like an hour and a half long conversation and turned around. And from that, grabbed the questions.

We have very specific laid out questions that were asked in the conversation, but with natural language, like we're having a conversation here, right? I say some stuff, we might go on a tangent a little bit. Things, aren't going to fall a very rigid sort of conversation. We're able to do with generative AI is basically take that transcript of a conversation and pull the answers out and put them into a document that can then be used later on as part of that company's process.

So, that kind of thing right there. Would have been impossible to do. I would say even five years ago. So just thinking about what that does for you. That would have taken people to do that. I have to now listen to the conversation. That's an hour and a half long. Take my notes, fill out this document, right?

And that's something that we could do in 30 seconds or less. You know what I mean? 10 seconds, whatever it takes just to go through that. But what that required was building out the logic and the code and the, the workflow for being able to ingest these transcripts and then take them and process them out and produce documentation.

So there's, that's the give and take, right? If there's an already prebuilt tool out there, it's is going to go into some of the things you're probably going to talk about, right? Is if there's a prebuilt tool out there that can hook into your system or do whatever you got to do, that's going to cost you X amount to utilize, right?

And, but if you need something totally custom, then it's all the development effort into building that out and building that in addition to the cost of, running AI and making those queries to Vertex or whatever, backend AI models that you're using.

So anyway, those are some of the ones that I've really liked, real excited, real fun to build, real excited to see the client's faces when they're like, this can be done? That's crazy. You're, you've just saved me days of work a week.

So that's my thoughts on what I'm really proud of building.

John Pettit:

So, we've also used it to extract information just from normal sources. So when you have a form and it's static and you know where all the fields are. It's one thing to extract the data.

You can do that without heavy AI, but when you have variable forms of how information structured, say you're going out to competitors' websites, and you want to see what's new in terms of products or feature sets, and you want to keep your database refreshed of that. You can't do that as easily.

But you can with a combination of, like a retrieval augmented generation approach where you're taking data from live sites, and then you're using that to be processed and summarized by the AI for specific factors. And like Justin said, one of the cool things about like Gemini is you can get back a JSON response. You can tell the AI, I want a JSON response that looks exactly like this. And then you can make it essentially an API to think programmatic.

But one other thing I think is neat about the, should I build or buy it? A conversation is some things are much harder to build than others. Like building a, your own kind of rag requires like embeddings, database, a data warehouse, a query processor, a data processor, a way to display and present that information. It's a pretty hefty application, and there's things out there that simplify that.

So Google has Vertex AI search but there's also tools out there like lean that are really powerful that can, you can implement, put on top of your data and then, You know, make it easy for all of your employees to find critical information. Or it could be consumers, if you wanted to put that out in a consumer interface.

Brandon Carter:

Let's dig into that a little bit, because that's a good transition. Last month, we talked about employee tools and how every company really needs to consider: do I want my employees bringing their own AI tools to work? Or do I want to give them something that I've managed in something that maybe has a little more boundaries around it?

I think the same conversation may apply here when you're talking about from a development perspective.

Correct me if I'm wrong, you can cook all this stuff from scratch, right? That's, I think that's what you were just talking about, John. But there's also a lot of pre made tools.

I don't know what you guys recommend for the typical organization. Like, the organizations that are in the middle, not a giant enterprise with an entire development department, not a startup where they may not even like they may have a developer, but for your common company that doesn't have infinite resources.

Do you, is it worth constructing this sort of back end, this infrastructure yourself? Or should you really look at bolting into something that already has parameters around it that's already built? What are your opinions on that?

John Pettit:

I, like nine times out of ten, I would probably say use something that somebody else is building and supporting. I think the cooking analogy is not a great one for this type of thing because it's not like I can cook it and it's baked and it's done and I get to enjoy it.

It's probably more like a garden. We're like, I have to plant something, but then that garden needs constant maintenance, right? Keeping the weeds out, the pruning, same thing with anything with data and AI is the data was trained at a certain point and life changes around it. And you need to take, change and monitor that and respect that.

And then just anything with infrastructure. Once you have a piece of software, like you have to patch it, OS is changed, compatibility changes, you have certificates, you have a whole set of infrastructure around it that you have to then own and manage, which is just general IT knowledge, not specific to the AI.

So, if you can leverage somebody else's thing, you should do that. And then if you're going to try and build it your own, you should probably go at it with a partner who's built and deployed something, right.

And there's the whole concept and the software development world around DevOps, which is like the operational side of development. There's also MLOps, which is like the data and operational side of machine learning. So you need to bring in. People that can help you not just get it running, but teach you how to manage it.

Aaron Gutierrez:

Yeah, that stuff's all on top of the the actual coding part you have to learn as well. So it's not just hopping in and, starting to write code right away. Yeah, John mentioned all of the infrastructure part and kind of tackle you before you get started.

Brandon Carter:

I'm curious to dive into that.

Like, all right, I've got a developer here, like he, he or she's got a particular skill set, how, what do you recommend for someone that like, I need to learn AI? Even if I hire out, a partner, even if I hire, or even if I bring on like a prebuilt model, still need to understand it. Like, how do you guys, how did you all get into it? And what do you recommend for the typical developer that's looking to get in there.

Aaron Gutierrez:

There are a lot of ways to get started. It depends on where you want to get started.

Do you want to be down in the weeds with the math and the machine learning like concepts that like hold everything up? Or do you just want to get in there and, build stuff?

I guess starting at the back of that, like something like a chat bot, like that's got so many new parts, you wouldn't know where to begin, but you can just use something, like Google has rolled out a new agent builder, which is pretty much like a, out of the box solution to get started on building a chat box, chat bot.

Before they used to be pretty complicated where you had to design all the paths and decision trees and everything. And even that was a little difficult, but like now it's to the point. It's pretty sophisticated.

You can just set up a new agent, depending on what vertical you want to focus on. They have like healthcare specialized agents. They have a code specialized agents. You can select whichever one fits your use case and just get started with a really quick demo that, does what you need for 80 percent of the way. But from there, you'll have to do a little bit more research on feeding it up to date information and all of those technologies are starting to get baked in as well.

Even the RAG stuff, you don't have to, you don't have to build a specific RAG. You just point it at data sets, you just point it in a website, and it'll pull real time data.

One cool feature I saw, you can even use custom functions now. That's crazy that you can have a bot that will make a function call and do whatever processing data needs to happen, and it will return it within the conversation and keep it in context.

But, there's, I guess to answer your question, you can start At the simplest, let's pick a simple bot, or you can go into the theory of it. Also, just depending on where you want to get started on that will be the answer to that question.

Brandon Carter:

Justin, I'm curious to get your thoughts here. How did you get into this?

Justin Barone:

So I'm like a, I got a, do it. I gotta get my hands dirty. I gotta get in there and feel the, kneed that dough and feel it for me to be able to, really grok something. So I would say, there's tons of online courses and different things out there. POCing out, just come up with a dumb little idea and try it.

That's basically, I'll say this. It's a little secret. Some people know this, but my secret is my entire career has been going and taking new technologies, taking a little idea, and trying to make something with it. Just that's what's fun for me. And that's how I learn and how I how I grok these technologies and download them into my dome.

From a standpoint of if you want to go open source, there's Hugging Face out there. Hugging Face is a really great open source community and kind of directory of open source models out there that you can take.

Some are too big to run on your local computer. You usually need a GPU to run them. So like an NVIDIA GPU, so I'm a big gamer, so I have my gaming rig, so I'll, I'll pull down something. These GPUs only have so much memory. So you can only run kind of certain size models on them.

You have the ability to go fire up a VM in GCP, right? And you can fire up a VM that has a big GPUs on it. It will cost you a lot of money, but if you want to try out some things or proof of concept, some things out, you can do that for an hour and then shut it down, right? And it only costs you a dollar, a couple dollars or something.

There's hugging faces also integrated with other kind of back ends that will spool up some of these models on the fly to play around with them. And then, thinking in sort of layers of abstraction you can go use the APIs, the Vertex API, depending on whatever chat or GPT models or whatever models you're using out there.

Like, the main ones that are closed source, they all have APIs. Those APIs cost a certain amount of money per, tokens. And that's how many words you're passing in and getting back, okay.

So you can play with around with that. What's really cool, Gemini and Vertex, they all have very very nice well-defined kind of starting areas that you can play around with and it doesn't cost you a bunch of money. So I like getting in there, building something real, getting my hands dirty.

The other thing is if you go to conferences, hit up all the ai sessions where people are talking about AI. Join communities. I'm sure there's Discord communities out there. There's communities all over the place.

And then also shameless: Promevo, we do workshops. We'll do workshops around AI stuff, especially if a client has an idea of something they want to do, but have no clue of where to start or where to go. We can help with that.

We can also set expectations, right? So like what John was saying before. You, what we were talking about having to build, have these dev teams to maintain and keep these things going forward. We'll help you understand like what that will take, and whether you want to go down that path or not.

So I think, and I think a lot of, clients or people out there that are interested in AI don't know where to start. It's, it's one of those things where you got to have someone who's done it and come over and help you. And you read stuff online. That's all great. But when you really want to, you need help with that. You can hit us up or, there's other people out there.

There's YouTube videos. There's tons of YouTube videos on AI and even getting into what, Aaron was saying about the like math behind it and how it works under the hood. That might not help. You're not probably not going to go build your own model from scratch. But that gives you a good understanding of where, where it started and how these things work.

So, I know it's a big long winded thing, but I'm like getting your hands dirty and like John's analogy getting in the garden and yeah hands in the soil.

Aaron Gutierrez:

Speaking of that. I forgot to mention like there's probably a lot of partners watching this the Google Partner Platform has a ton of labs that you can go on and use their training. And that's really hands on and that'll, that'll take you.

John Pettit:

You don't have to be a partner for that. The Google has Cloud Skills Boost.

Aaron Gutierrez:

Yeah.

John Pettit:

So anybody can access that. And if you wanted to give a subscription to a couple of your employees, they could use the lab time on Google's resources and not play with your stuff. So yeah, definitely good walkthroughs in there.

Justin Barone:

Join a hackathon.

John Pettit:

Join one or create one, right? You had the concept of a Promevo workshop, maybe have one of our engineers help lead a hackathon with your team.

And we sent our people off to these things as well. I know Aaron, you were just at an event off in Chicago with Google, but we're happy to help, spread that knowledge, to help make your people. What are the L 400 AI wizards or whatever the new leveling is.

Brandon Carter:

You guys touched on a lot there.

You actually naturally segued into several more of the topics. I know we're starting to come up on the time. I do want to talk about what an engagement with an AI partner like Promevo looks like, which we touched on there.

I think before we do that though, let's talk about Vertex and because I've heard each of you mentioned it.

We are obviously a Google Cloud partner. So we work in Vertex a lot, but tell me about like the idea of what is Vertex and how does it help you as someone looking to integrate AI into your products and services?

John Pettit:

So Vertex is Google's brand for all their AI. And what it is they're managed platform for managing your whole AI infrastructure behind the scenes.

So you can spin up AIs from Google or other third parties. It could be open source. It could be your own models, but managing through one central interface and have a consistent set of APIs to access it, measure it, monitor it over time. And that's the key benefit there. It follows Google's desire to be, platform agnostic, right?

Whether you want to run stuff that's theirs or something else, or you want to have it portable to another cloud. It still allows you to maintain that level of freedom when you're building your applications.

Brandon Carter:

Gotcha. So in preparation for this, myself being not a developer, I'm the marketing guy. I need to know a little bit about it, but I don't need the level of expertise that you all have, but I did actually I asked Gemini what is the difference between Gemini and Vertex?

And it actually gave me what I thought was a really good response. It said, Hey, think of Vertex as like a construction crew, who's building a home and building your architecture and like building this framework for how something's going to work. But then think of Gemini as I don't know, like a smart home assistant like Google Home, or I'm not going to say the name of the Amazon one because it'll start talking. But like something that's already built, it's like it's prepackaged it, it can be evolved a little bit, but it's really meant to be like something that you turn on.

And correct in saying I thought actually that was a pretty good analogy. So one resource for a developer is to ask the AI itself to explain itself.

But yeah, so Vertex is essentially, is it fair to say that it's like a shortcut of sorts? There's a lot of pre packaged things there that you don't, we talked about building something from scratch.

Vertex is a shortcut to that. Am I, is that an accurate phrasing?

John Pettit:

I think it's like a toolkit that allows you to assemble your ML and deploy it. Um, instead of you having to do this by hand and spin up the infrastructure, the things we talked about earlier. Like certificates, API endpoints, all this other stuff. You can use Vertex to take the components and spin it all out together and manage that over time.

Aaron, do you have a better analogy?

Aaron Gutierrez:

Oh, for me, for things I like to do, it's just, it's a way shortcut because I don't have to mess with my Python environments or anything like that. Like I can spin up a notebook, start writing code right away. So yeah, I think the shortcut. If you're a Python coder and you want to get in and start, making POC, you got all the models there at your disposal.

You've got so many resources and you don't have to worry about installing too much. You've got a. Actually hit the libraries you need, but that's the single line of code in the notebook. But other than that, like you're not having to worry about the infrastructure part itself, which is really nice.

Justin Barone:

Yeah, I like the toolkit analogy. I think it's, that's a good example of it because it's there to call upon the Gemini Pro model or, the different models out there. But it allows you to call those using the same kind of patterns and practices, making it a lot easier as an engineer to be able to implement those things.

John Pettit:

Gemini is just one type of AI model, right? Like it's a language model, which is great for interfacing and getting explanations and summarization and all this stuff.

But there's vision models, there's text analysis models, there's text to speech models, there's all kinds of audio models, video models, different things that you could plug in through Vertex to solve a specific problem in your business to automate.

Brandon Carter:

Yeah, I think that's a really good point. And, obviously as a Google Cloud partner, we work with all of these I've got a list over here of Gemini Code Assist, AutoML, Document, speech to text to speech, translation AI, there's a lot of different AI tools, like Gemini.

Anyone that watched any of the Olympics, you saw Gemini, like you saw the commercial with the Jay Z song, like it's out there and a lot of people are adopting it and using it, but that's just like the tip of the iceberg for what Google provides and for specifically what Promevo can help with.

So I want to revisit back to that because, obviously Promevo, we do like GCP work. We do a lot of Google Workspace work, obviously gPanel, which I talked about at the beginning, but talk to me about what is the typical engagement from a client that is exploring AI or wants to use AI? What does that look like? You mentioned workshops and ideation earlier. Is that, what is a common engagement look like on the Promevo side?

Justin Barone:

So really, it's we have a, an initial consultation with them, want to understand their business needs, their goals. We look at do they have people on staff that are developers or ops people? We're really just trying to understand their infrastructure, their AI readiness.

Are they ready to do something custom? If it's going to be a custom. The development of the custom solution is usually going to be an iterative kind of testing deployment process, engine development, test deploy real similar to any other kind of software development.

And then, once we have a solution in place it'll usually be ongoing support, optimization, client feedback, roadmap on what they want to do in the future. That kind of stuff.

So it's, it tends to be depending on what they're looking for. It can be very different and the outcome. Sometimes it's, Hey, we've got something we started and we don't know what, how to take it. We got something we started. We don't know how to implement it in the cloud.

We have an idea. We have no engineers. How do we do this? Is it possible? Does it make sense?

So it's really just really understanding their needs and their goals, and then going from there and building almost a custom engagement around that.

John Pettit:

From a business standpoint, since we're a Google partner if you're looking to leverage Google technology and maybe you don't have much cloud spend or any cloud spend, we can also help work with Google and you to create, programs to help offset costs and make it affordable for you to get started in the cloud. That's the benefit of working with a partner instead of just spinning it up yourself.

The other thing that we can do is meet you, I think, to put Justin was saying where you are with the problem. Are you just wanting to know about AI, we can do a workshop to teach you about AI and what you can do with it. If you need help identifying your problem, we can help you do that.

But if you're ready to get going and yeah, we can do a development or kind of work with you or for you to build out, whatever it is you need to get done.

Brandon Carter:

I'll drop in another marketing pitch here where probably, I don't know, probably six or so months ago, Justin and Aaron did a webinar on Google's serverless software development. And one of the things that you guys talked about is you presented a case study of a, a client's tool where their bandwidth had run amok and their costs were going up and down. And you talked about how you streamlined it and made it more responsive and reduce their costs. Because one thing that you guys didn't mention is like, Hey, we tried to build AI into our app and it went really poorly.

So that's something else that we can do. And there's a pretty good case study that Justin and Aaron talked about that we did live just, I don't know, probably six months ago.

We have a few minutes left. The last thing I want to talk about is like security and ethics and like, how do you make sure that you're not like somehow exposing your data to the world or, maybe a potentially malicious bot?

Just real quick question. What are like the cost pratfalls without maybe getting into like specific numbers on because it always varies, right? Like what it might take, what we might charge, what someone might charge to maybe develop an app or fix an app.

But what are some of the like costs like pratfalls or potential dangers that people need to be aware of before they get into this? Like, where could you lose a lot of money, or where do you see people potentially burning a lot of money trying to get AI into their tools?

John Pettit:

Real world examples that i've seen out there with customers or previous places I've been is when you're going in to do training and not having a really good process yet of how you're managing that training run and set, you could quickly run up a bill of, $ 100, 000 or more, right?

So maybe you just went in there because you wanted to fine tune something in your model. And you're taking full ownership of that model and you're not really using foundational stuff. If you're not really clear on how you manage that gate process, yeah, you can go wild, which is another case for why you should be using someone else's model or API rather than trying to spin it your own.

Cause even really smart people I've seen make that mistake and then have to come in with their tail between their legs and talk to their boss about, I made a big mistake and do I still have a job? That is one quick way you can have a big mistake with it.

Justin Barone:

Yeah, and I'd also just say that, the things that you care about are data privacy, compliance, regulations, right? GDPR, HIPAA.

You have if you're using AI the business is going to have a policy on what you're allowed to put into that AI, right? So if, we're using AI and ,and I do know on the enterprise level and John, you can probably talk more about this on the enterprise side of Gemini, for example. You have those controls and kind of those levers to be able to constrain whether the AI can train off of the things that people are inputting and that kind of stuff.

I think those are really important, and it matters at the business level. So when you get to that Hey, we want to put this out across the whole company and let everyone use Gemini. Those tools matter It's not just bring your own and you know someone over here is using anthropic and someone over here is using Chat GPT and and they're feeding it information that's like proprietary, NDA business stuff into these APIs or these other AIs.

I'm curious, John, though, your thoughts around that, especially from the administrative.

John Pettit:

Yeah, every, everybody you share data with, you need to make sure that you understand the data processing security and how they're using it. And so having just sprawl out there is a nightmare probably for a CISO, unless, people are trying to bring their own AI into the situation.

So in our previous webinar, I know we talked at length about that. And I think it's important. People can go back and look at that, but it's make sure you have policies, make sure you're keeping it secure, make sure you understand what tools people are using.

Building out your own AI, you have to consider what you're building on top of. So IP, do you own the IP or are you delegating IP off to someone else? Is that data that you're pushing into it being used somehow to train some other foundational models? So you need to understand where the data is going.

And then Justin, you just brought up that point of somebody's using Anthropic, somebody's using this over there every time you're trying to get data from point A to point B, like you have to keep that data secure throughout the life cycle, and you probably should redact it using something like if you need, if you can using something like the DLP API from Google or something like that to preserve the data integrity and shape, but not leak important details that don't need to be everywhere, right? So keeping controls around what data is flowing into one APIs for what training sets.

Again, like the painful recent, like 2. 9 billion social security number hack. Somebody somewhere probably had a file out there of all this information unencrypted, right? Or on a database unencrypted. And for whatever reason, or whoever they were sharing it with.

Because these breaches, for somebody to pull that much data, it's likely somebody made a mistake. So with APIs and AIs you have, or AI tools, you're going to be sending data in bulk in different places to fine tune models. Make sure you have a process for governance around that data and security and encryption and redaction.

Justin Barone:

And you also mentioned I guess I brought it up earlier and you said you wanted to come back to it, the sort of ethical concerns. And that's right. I think from that standpoint, it's am I cheating at work by using AI or, you know what I mean?

Like you, I would say when AI first started getting wild and I was using it, I was a little bit like, is this cool , are we, and it wasn't until my boss John up there was like, Hey, we should be using this stuff, but this is how you use it. And these are the policies that matter that, that you have to follow.

And this is the AI that you can use and kind of those things, but there is that little weird guilt, and I'm sure someone had that guilt when the calculator came out, I remember being in high school. Cause I'm old and them going. You can't use a calculator on the test or whatever, right? And or middle school or whatever. But, there's that whole thing of there, there's this little, there may be this weird stigma as everyone's getting used to this new technological shift.

John Pettit:

Yeah, I think that's very true. I know we, we've tried to make sure that we let people talk about how they've used it successfully and share it. I think it's important for people to do this out in the open in their companies and not shame people if they're using AI to do their work and, not wonder are they really producing anything or could I replace them cause they're just using AI?

I think if they're doing it correctly and creating good content. People should celebrate that. And they should share stories like you did on your video, Brandon of here's how I was able to create a marketing plan. I think that a great way for everybody to win and not create like this semi uncomfortable culture of is it safe?

Ethics is a harder one around AI. Security is one thing. Ethics is a much harder philosophical question about what is ethical and not to do an AI. And Google's taking their stance, Microsoft's taking their stance. Everybody has a stance on how you should be protecting people using AI from doing something malicious, hurtful, hateful, things like that.

And I would say that instead of spinning your own AI, using a foundational model is probably a good bet so that you're not creating your own stance on that as a company. And like you're inheriting one that you can, then build on top of.

Justin Barone:

I'll give you a good example of that is the writing community.

I like to write short stories. The writing community is like anti AI all day long, like almost like a hatred for it. If I write a short story, I'm going to pop AI and go, I want you to be a critic of this and tell me your interpretation of this right and I'm not doing it to cheat my way into writing my story.

I still have my concept and my plot and you know everything that I want to convey. It's just if I go and ask an editor to look at it, but how's that any different? It's just an editor that's not a real person.

So there are those, I think, still kind of stigma things and you know for the sake of art is art right art made by humans. There's also art made by So it's one of those very interesting things.

Brandon Carter:

I think that's a really good point to go out on. I think if I can summarize or attempt to summarize like AI, we're moving out of that wild west era, like initially a couple of years. I know we just hit I just saw we just hit like the 26th or 36th anniversary of when AI really became a thing.

Like we're talking all the way back to 1988. Then all of a sudden, what, two years ago, like Chat GPT comes out with a free model and there's this massive explosion of DALL-E, people generating weird images that went viral online, voice imitation. And that's still there. But now we're starting to see things take shape, right?

We're starting to see it in the business world. Last time we talked about last month's webinar, we talked a little bit about is there a need for a Chief AI Officer in your organization that there may be, for some organizations that may be the need, but I think relevant to what we're talking about today is all right, this is taking shape.

We're beginning to know what we didn't know back then about how to put it in your organization, how to put it into your products, how to make sure that you're not like burning through, like exposing your secrets to some malicious actor, and I think that's where again, like just not to make this a sales pitch, but that's where companies like Promevo come in and we help you with those, put those guardrails around your AI efforts and help you do it right and do it correctly and do it in a way that enriches your users, it enriches your employees and ultimately enriches your company.

With that, we're at time. I'm going to put one final pitch in for next month. We are going to be talking about AI adoption and basically some of what we touched on today.

But you're spending a bunch of money as an organization on AI; how do you make sure your employees are using it? How do you make sure it's being used properly within your products and services?

So we're going to dive into that exactly almost one month from today, September 17th. So those of you that have registered for today, you're already registered for that one. We'll send you an email letting you know where to go to watch it.

For those of you that are watching this on demand, come and sign up. We'd love to see you.

With that, I want to thank John. Thank you, Justin. Thank you, Aaron. Much appreciated. Great conversation. Good thoughts. Thank you to our viewers, both live and in the future. Much appreciated. We will see you soon.

And in the meantime, have a great day. Bye now.