#125 - How AI and Automation Is Going To Be Essential for Business with Andrew Hewitt I The Corey Ganim Show

#125 - How AI and Automation Is Going To Be Essential for Business with Andrew Hewitt I The Corey Ganim Show Speaker 2: Welcome back to The Corey Ganim Show. So this week we have a repeat guest. We have Andrew Hewitt of YoDev. Now, some of you, if you've been listening for a long time, I'm talking like maybe since day one, you might recognize Andrew's name. We had him on back on episode 16, which was September 6th, 2023. So over two years ago at this point. So I encourage you guys to go back and listen to that episode if you haven't yet. And Andrew is a very smart guy when it comes to development. He's very technically minded. So I wanted to bring him on to talk about AI. So some of you might know I've been getting deeper down the AI rabbit hole, really exploring a lot of things as it relates to AI and automation and just really getting into that world. So I wanted to have Andrew on. To talk through AI and automation now, the way that we're going to structure this episode is we're going to start off at a little bit of a more basic level. And throughout the course of the episode, we're going to get slightly more in the weeds throughout the episode. So if anything doesn't make sense, if anything, you know, you want us to go in deeper on, on a future episode, feel free to leave a comment if you're watching on YouTube, or you can DM me on Instagram at Corey Ganim if you enjoyed the episode and want to see more. So Andrew, thank you for coming on. I'm looking forward to our conversation today. I think we're going to really be able to nerd out on some of the stuff that you and I have been talking about for a while. Speaker 1: Yeah, Corey, it's a pleasure. Thanks for having me. Unknown Speaker: And I'm super pumped to get into the conversation because I know you and I have been trying to connect now for a while and we're just so busy. So this is great to do it here. Speaker 2: Yeah, looking forward to it, man. So let's start off by really quick at a high level. Why don't you just give a background, you know, 30, 60 seconds, your expertise, just to kind of set the stage for your knowledge. Speaker 1: Sure. So, um, the accelerated version is, uh, in 2017, I transferred from being a sales person for like a really fast growing startup company. To going out on my own as a marketing consultant, which then led to me niching down on just web dev services. And then once we niched down, like the riches are in the niches, right? So like, it just took off. And then I started hiring team. Now fast forward 2019, or 2019 is when we like really started kicking off and then like, Like rapid growth now, 2025, we have a leadership team, engineering manager that works alongside of me that's way more experienced than me in engineering. And along the way, I self-taught myself programming. I got to work on contracts with big brands like State Farm, Wells Fargo, doing software development at a really high level before AI was a thing and everybody could kind of dabble and get things done. So I have all that knowledge. And currently right now, YoDev's scaling our full stack engineering services and like our DevOps and higher level consulting services for midsize and enterprise level businesses that need not only software engineering and support, but also now we're getting into AI consulting and implementation at a really high level. Speaker 2: I love that. And yeah, that's why I wanted to have you on because you understand this stuff at a much deeper level than I think probably 99.9% of people do, myself included. Now, my audience for the most part is non-technical, right? And neither am I. So that's why I wanted to, like I said in the intro, start From a high level and get maybe progressively more in the weeds as we go along. And I know you've got a pretty cool demo to show us at the end of the episode of a custom GPT that you've built and showing people how that can fit into their business model, right? Whether they're selling on Amazon, whether they're running any kind of business, this is the type of thing that really will benefit you as a business owner, especially if you have a team and you're actively hiring people. To start off, Andrew, at a high level, right, I think when most people hear AI, they think ChatGPT, right? And then the people that maybe are a little more advanced, they think Claude and Perplexity and some of these other platforms as well. But 90% of people think ChatGPT. Again, from a high level, if you had to just break it down to the everyday person, what are the things that AI can do for us today that we couldn't do, say, even a year or two years ago? Speaker 1: Yeah. I mean, I'm sure many of your audience is already using AI for these things, but just simple things like, streamlining, busy work, like, you know, proofreading, you know, your your messages. And by the way, I saw your LinkedIn post, I'm a big advocate of like, Keep writing your own stuff and use AI to get better at it. I found myself actually, I'm a way better writer now. I typically now, instead of just relying on AI to constantly write my emails, it's made me better to where I can just off the top, just boom, just knock out beautiful sentences. I'm right there with you with that statement that you made on LinkedIn. It can help you with that. It can help you generate proof-of-concept images that you can easily hand off to a designer that like, you know can save tons of time there. It can help analyze, a lot of our clients are using it to analyze marketing data and KPIs and trends. So like connecting things like Google Analytics, Fathom, your meeting notes for like your sales meetings and prospects or whatever else, like you can kind of connect all that to either your account through connectors or creating custom GPTs that stream that data flow into that so that you can analyze and run your reports without having to manually go through them. And then a little bit on the deeper side of things, it's getting really good now for on our end of things with coding. So, you know, cloud code, things like that, really can anybody with like, you know, probably just like base knowledge could probably spin up some pretty Nice things. You'd still probably need somebody a little more technical to take them the end of the way, but you'll save so much time and money just getting that proof of concept to hand off to your engineer or your agency to like, oh, they know exactly what you want now. There's none of this discovery and back and forth that's just burning time and money. So super like big efficiencies in project management too. So like almost all the major project management tools like ClickUp, Asana, Jira, probably Basecamp and whatever else. I haven't been on those. They're integrating like an AI brain, essentially, like their own custom GPT that's scoped to your workspace. So like you can just hit your AI brain and your project management tool and then just spit out summaries of projects or where statuses are and things like that. And then as we go even a little bit higher level, I'm not going to get too much into the weeds yet, but now with all these tools having the ability to use AI, all of them have an MCP server ability. MCP, for those of you who don't know, stands for Model Context Protocol. It's essentially how your tools allow AI to interact and connect with the data that they're storing. So for instance, if you got Jira or ClickUp, and then you got Gmail, Google Workspace, and then you got XYZ, they all have protocols where you can connect your custom GPTs or your custom apps that pull from and access that data. So now instead of having just a ClickUp brain or a Jira brain or a Gmail brain, now you have like your company's brain that's then connected to all your third-party tools that you're using. And we have proof of concepts of that working and it's really, once it's buttoned up, like it's phenomenal. All of our most important tool sets are going to have like a YoDev MCP protocol that connects to all of our Most important tools, and then we're prompting it with our workflows for onboarding new clients, delivering projects, et cetera, so that it can almost automate a lot of the setup of projects, client drive folders, all this stuff that you would typically use like make.com or whatever. We're figuring out how to do it just through the APIs and the LLMs and things like that. So it's really... It's a really exciting time to be alive right now, especially in our industry. Yeah, it's really fun. Speaker 2: Yeah, the MCP thing is really interesting because I guess the most basic example that I found and kind of what made it click for me is in like a Claude, for example, right? If you're using Claude, you can connect your Gmail to Claude. A lot of people just see it as like, it's called a connector in Claude, but really what it is, is it's an MCP server on the backend. And you can go into Claude and prompt Claude and say, Hey, what emails have I received today? Right. And really by, it's just what gives Claude access to your email. So Claude can go in and say, well, Hey, you've got these 10 emails. And because it's connected to Claude via MCP, you can actually prompt Claude to then do things in your email. It's not just going to read them to you. You can actually respond to an email by prompting Claude and saying, well, hey, you know that email that I received from Andrew? Why don't you go ahead and shoot him a reply and tell him I'm interested? And you can do all that from within Claude, right? Same as ChatGPT. That's for a while. I kind of struggle to understand like what does MCP mean? But that's the most basic example that I think made it click for me. Speaker 1: So yeah, it's um, it is wild and Like we we working on like everybody here probably knows like or maybe it has heard of WordPress before but mm-hmm like There's no out-of-the-box MCP for WordPress But I was just playing around with it and I created a little MCP server for WordPress sites and you can use Claude code to connect to your WordPress site and it can draft, it works. It'll write blog posts and then it'll publish or draft them up there on your WordPress site ready for you just to review and then schedule out or you could schedule it and post it right there too. The actual building out, The beautiful landing pages is kind of hard. Like it tries and it's like all jacked up, but like the content's really easy because it's just formatted content in the content area. So yeah, that's another example that like I play around with something custom with MCP, like creating that custom like WordPress MCP that connects to my Claude code CLI tool or that stands for command line interface. So like right in the terminal, but it's really cool to like, it's awesome. Speaker 2: Yeah, that's just a lot of the integrations that are already happening, but the ones that are in the pipeline too are going to be game changer. I think that's what's going to make this technology that much more accessible to the everyday person is the integrations with all your tools, right? Because again, right now, most people just go into ChatGPT and ask a question. But what happens when you can go into ChatGPT and tell ChatGPT to send an email or to schedule something on your calendar to reschedule this thing or whatever that ends up being. So that's all here. And if it's not here, it's on the horizon. Speaker 1: If you're interested, you can just search MCP. Repos, GitHub or whatever, and it's just like an endless list of every single tool you could ever imagine. It might not be directly in the connector for ChatGPT, but if it's in a list and there's a repo for it, you can connect, you can spin it up a little custom. So there's endless MCPs out there now for every single tool you could think of. Speaker 2: And in spinning it up custom, right, that might sound intimidating to most people, especially me. I'm like, I'm not technical at all. The thought of having to code anything is just so outside the realm of my expertise. But I realized that connecting. MCPs to like Claude, for example, for a tool that doesn't have it natively integrated. It's super easy. It's like you're copying and pasting like two lines of code, at least the ones that I did. Yeah. If I can do it, it's doable. Trust me. Speaker 1: It's not hard. It gets hard when you like. You know, when you're in a situation that we that we typically get in where you're larger business or enterprise and they have a lot of red tape and like you have to actually build something and like scope it out because they don't want all their data accessible and you got to build something custom and you got to put custom endpoints in there. But like for stuff that you do like You should be able to easily just use the AI to teach you how to connect the MCPs that aren't available in the connector and then it'll just walk you through how to do it. It's not hard. Speaker 2: Right. Yep. Now, my next question for you, Andrew, so what are your top three either AI tools or AI assisted tools right now that you're using and why are they in your top three? Speaker 1: Yeah. So I never strayed away from ChatGPT. So, ChatGPT is something I still use every day. And I'm using it in like a different level now, I'd say, because I am using the connectors and I'm using some custom MCPs. So I'd say, you know, from the average user, I'm getting a lot more out of it. And my personal user's context profile is so good. Anything I need for YoDev, higher level things that I'm doing, it just already knows it. So I can just, boom, knock things out. So I'd say ChatGPT is still number one for me. I'm a big advocate of I adopt a tool and then I just grow with it. I never had any issues with ChatGPT at all, so I really didn't have any reasons to leave. So ChatGPT. Now, my tools are probably going to be more coding focused. But Claude and Claude Code is becoming a solid number two. Speaker 2: Yep. Speaker 1: I like Claude's UI. It's nice. I think it can do just about everything that ChatGPT can do. But Claude Code is phenomenal. Speaker 2: That's what I've heard. Speaker 1: It's one of the best things I've ever, like it is just awesome. So to put it clearly, like, you know, most people, Maybe if you don't know what you're doing with coding or even if you do know and you don't understand that there's better ways to do it, you work in ChatGPT and you work through and it gives you snippets and it's almost like following a tutorial and then you go to your code editor and you paste. Well, with Claude Code, it's a CLI tool, which stands for Command Line Interface. That is right there on your local machine with you. And then you can actually change the directory of the code that you're in, and then type in the command Claude, and then a little prompt pops up, and then you just work with it just like you would up online on the interface. And then it'll go through and make the suggested code changes, and then you can go and approve them, and then it just changes it right on your file system. It just goes and changes multiple files and then you can test the code right there locally. If everything's good, we push it up to GitHub and boom, that features out the door and Claude's doing the heavy lift in there and I'm kind of just like a senior engineer that reviews the code. So Claude code is definitely number two. Our efficiencies are going through the roof with that. And then number three, um, I would have to say, uh, it's not really AI, uh, but I'm, I'm starting to leverage a lot of the low code, no code, um, tools like make.com and things a little bit more. It's more automation. But there are some AI features baked into it. Speaker 2: Right. Speaker 1: And it makes it easier to connect to different artificial intelligent tools and things like that that can help automate processes while incorporating AI at the same time. So we're using a lot of make.com to streamline workflows and proof of concepts internally. If we have employee onboarding forms or new client intake or whatever, a lot of that heavy lifting on the process side and even connecting AI agents to give summaries or whatever is coming from like make.com is our preferred one there. Speaker 2: So what are some of your favorite automations you've built with something like a make.com? Speaker 1: I like, so I have a Google form that my executive assistant fills out. We could also automate this even more with like just letting clients fill it out, but she fills it out when a new client comes on. And then make.com sends a webhook to Google Apps Script. And then Google Apps Script, um, does a lot of, so like, you know, me being a developer, like it's a little, it's easier for me to, um, you know, like kind of just working Google Apps Script. Speaker 2: Yeah. Speaker 1: Apps Script is like, does a lot of heavy lifting on in drive. So what happens is Google Apps Script goes and puts our client folder infrastructure in place perfectly. And then once all that stuff, once that job's done, make.com fires off again and hits our template set for like all the different templates that we need for each folder within our drive folder structure. And dot com will then go populate the templates with the client information and just populate all that right into the, Yeah, right into the folders. So there's a couple steps there and there is a custom component because that Apps Script thing is really cool. It not only sets up the files, But it also sets permissions for my team. So like leadership gets a certain folder, like the admin folder, and then the dev team gets the project folder. So that they automatically, like whenever the projects actually get kicked off, everybody has access. Speaker 2: Right. That's like, see that's saving a ton of time. Speaker 1: Yeah. So like, yeah, but so make is a, not the biggest lift. I'd say Apps Script's doing the big list, but makes doing a lot of the handoff webhooks to like get things going and trigger the events and things like that. Speaker 2: Right. No, and I love, I love the tools like a make.com or like a Zapier. And I've been tweeting about some, some automation, just some basic automations that I've been messing around with lately. But one that I built recently, and this was with Zapier, not with make because Zapier has a native integration with Fathom and Fathom's the note taker of choice that I use. So I set it up to where anytime a Fathom call recording is completed, Zapier will take that transcript of that call, feed it into OpenAI, essentially feed it into ChatGPT, and then extract anywhere from two to five content ideas from that call. Right. Because what I was finding is that there's so many times I'm having conversations, whether it's with my team or just with anybody, right. Any of these calls that I'm on, there's content in those calls, whether it's a LinkedIn post or like a tweet or just like a lesson learned or like a build in public type post where it's like, hey, this is what I learned today. And I was finding I was just losing out on a lot of that content because I didn't have a way to capture it. As long as the Fathom call recorder is in my call, it's going to capture that. It's going to build a transcript and then ChatGPT is going to take that transcript and turn it into two to five content ideas for me and just store it in a Google Sheet. So then anytime I'm looking for, if I need an idea of something to write about or something to post about, I can just jump into that sheet. And there's basically, you know, tons of post ideas just based on the conversations that I've had that week. So they're relevant. They're timely. Yeah, it's I think that's a real and it was a simple one to build. It took like five minutes. And again, it's just in Zapier. So there's no, no technical expertise required. Speaker 1: That is awesome, man. That's a great way. You're already leveraging the AI notes and stuff just for regular. Then you're double leveraging it. You're set up something to also pull other low-hanging fruits out of it that you normally wouldn't be focused on or have the capacity to because you're putting all your time into the call. Speaker 2: Right, exactly. So, and kind of on that topic, I mean, Andrew, what are some other, are there any other automations you can think of that might be some low-hanging fruit that you see people, it's like, why are they still doing those things manually when they could be using something like a Make or something like a Zapier just to get it off their plate entirely? Speaker 1: Yeah, I really think like I can speak a lot to just like delivering products or services like just the AI like you know all the AI stuff and the automations that you can do around managing projects or deliverables. Managing your pipeline too is another thing like I mean HubSpot Their AI is amazing. I forgot about that. Speaker 2: Oh yeah. Speaker 1: It's so cool. But then you can connect HubSpot to these automations as well and at different trigger events. Before your meeting comes up, send your sales rep a little rundown of the prospect, all the touch that they had with your company. Because the HubSpot MCP, You can drill into deals, companies, or contacts level, or all associated with that one contact that you're meeting with. I've seen one where you can have a summary sent out prior to a call being booked or something. Let's say a prospect in your pipeline books a call. You can then have a summary kind of just sent out to you ahead of time that you can review so you don't have to really, you know, do the digging, the LinkedIn digging and like doing the prep work for the sales call. You can just have like a little summary sheet in their prospect folder already. So that's one that I played around with that worked. I'm still kind of ironing out the kinks because like I'm starting to scale my sales process now. So like I'm starting to get into that realm of looking at that and how I can leverage automation and AI in my pipeline. Speaker 2: That's so good because that's actually, I'd never thought of that use case, but think about how many different business models that applies to and not just if you're using HubSpot, but really any, any CRM that has a, like a make.com or a Zapier integration, which is pretty much going to be every CRM. I know go high levels are really popular one too, but I mean, if you run a business that where the sales process is based off of like getting on a call with the prospect to close the prospect, I mean, it'd be pretty simple. I'm talking like 15, 20 minutes to set up an automation where let's say your salesperson has to jump on a call with prospect Andrew at 4 p.m. Well, you could easily set up an automation that either sends them a Slack message or even a text message five minutes before the call with a one paragraph summary of all your interactions with that prospect to date, right? So it's like, hey, I've got a call with Andrew in five minutes. I checked my phone. I see that we've been in contact with him for the last two weeks. He's kind of on the fence. He's got this main objection. All right, good. I'm ready to jump onto the call. Instead of me having to stop what I'm doing, jump into the CRM. Scroll back through all the notes, read through all the texts, all the emails. I mean, if you're at scale, especially, that would be like a game-changing automation, especially if you have multiple reps, you're doing multiple calls per day. Yeah, that's a big one. And I mean, you could go and sell that automation to a company that is doing some of these processes manually for, you know, a couple grand easy. Speaker 1: Yeah, I also have a, um, and I don't know if we have time for this demo, maybe another day or when we see each other in person, but I created a web application that creates sales presentations based on, you know, all the meeting notes and all the things that I've gotten. And I have a prompt that does it. And then it actually will prompt the, the application to then build out my pitch book. And I call it PitchBook. And then it has the slides that I know need to stay there, but it tweaks, it modifies it just enough to make it seem like it's a custom touch. And then the pricing options and the timeline and the next steps, which is typically custom in my pitch process, those are custom to the project based on the requirements and the scope that we discussed. And then at minimum, so just kind of give you an idea, right? I had a sales pitch today. And I had a late night work session with a big because I had a big client push for Q3. There was no way in my old process that I would have got this sales pitch ready. I literally woke up at 645 in my bed. And I had a sales presentation ready in about five minutes. And it's beautiful too, by the way, it's on brand, it's interactive, it's a web app. And then after I get it ready, I can actually deploy it to a unique link. That only they can have and I can it's password protected. So I can make it seem like it's like, you know, a little secure. And I'm doing this now in five minutes, which used to take me like working in Canva, even with Canvas AI, it would take me like an hour, like minimum. Speaker 2: Jeez, that long? Speaker 1: It was just like, I'm not a designer. Going in there, I could get it done, but my wife, who's our brand ambassador, she would be at my neck about it, because I go in there and mess the decks up all the time. This is my vision for it. The summary is one piece. But now imagine you get a Slack message saying, hey, you have your presentation coming up at 4.30 with so-and-so. Here's a link to the prospect's profile. And also here's a link to your pitch book. Make sure you double check and just like look at it one last time before going. And then literally that's what it is. I click the link, I go through the deck and then I see one or two things that I want to change. I just tell the AI I want to change it and then it's done. Speaker 2: You can even take it a step further and have it schedule 10 minutes on your calendar, an hour before the prospect call. And it's literally just a 10 minute time slot for you to go in and review the presentation. And yeah, so that way you just don't forget to do it. Speaker 1: You have those links right in the event too, like the client profile and the link to preview your pitch. Speaker 2: All that is capable. Speaker 1: I'm doing it now. It's still a very manual process. I have to manually prompt, but I already have the infrastructure diagram that I need to make it an actual web app that can allow anybody to go add their branding, create their pitch book that they know works, set your prompts, set your context profile, and then just have your sales meeting notes flow through it and just have your pitch books just automatically created for you. Speaker 2: That's a product right there. I mean, think about any agency or service business that does custom quotes. I mean, most of them are probably doing things exactly the way you were before. They're manually putting together these quotes or manually putting together these pitch decks. And usually it's the owner, like you, right? It's the owner doing the work. And I mean, if you're talking about an hour or more per, basically per pitch, and you know, at scale, you're probably doing 10, 20 pitches a month. Of the owner's time, I mean, that is, that's worth tens of thousands of dollars, like not no exaggeration. Speaker 1: I've been taking more sales calls. And since, so since it's consistent and it's, and even better, it's trackable. Cause like it's all in GitHub. So I know like, like I can kind of go back and be like, and see like, okay, when we, when we, when I nailed down this flow of the pitches, our close rates went from like 50 to 70%. Right. So like now I'm at this version of my pitch book. I'm closing at a 70% rate. The effort to throw these decks out the door and be ready and make the client really feel like, wow, this person is engaged. I am because I can be engaged on the call and not rely on having to stress about putting the deck together to make sure I'm speaking directly to them and they feel the connection. It truthfully is a connection because I care about what matters and then all the other stuff's there for me. I don't have to think about it. I can just walk through and then like, yep, this is exact. They go every time. It's the exact same thing. It's like, that's exactly what we need. Yep, that's a good timeline. Like just constantly, like they're just hitting it and it's just, it's exciting to just, the closing rates are just getting better and better as the next iteration of my pitch book keeps getting developed. Speaker 2: Well, it's a two-pronged benefit, right? On the one end, like you said, you as a person can be so much more dialed into the conversation. You can be 100% present. You can build the relationship that much stronger because you're not having to worry about, oh, shoot, I got to take notes or, oh, I got to make a note of the exact price that we quoted them so I don't mess that up in the pitch deck. It's like you can be 100% present for the conversation. And then the AI is just going to deliver the same repeatable product every single time. And well, obviously, they're going to think, yeah, this is exactly what we need, because it's literally take the AI is probably taking their exact words. Speaker 1: Yes. Speaker 2: And just putting that into pitch deck form. Speaker 1: And so, I mean, I have to do is worry about making the numbers work, right? So then it's like, right, numbers work. Like that's literally that's the, mainly that's the most thing, that's the thing I got to change the most is the numbers. And then sometimes AI like, I'll say this, until I have my own app and I have my own like rag setup, which we can talk about rag later, but like until it all has its own rag backend, it's going to be prone to hallucinations. So like, it'll like, it'll, it'll throw in like, wow, that's a really cool thing, but I'm not going to promise that that's the top. It's like, you know, like until I have a sophisticated product with some guardrails and stuff, like there is a little bit of manual intervention, but nowhere near having to constantly just keep building the decks every time. So that's, That's something that I'm really excited about and you are correct. We are moving fast to make it a product and we have several agency partners locally here that are wanting to beta test it and also be a part of building it too. So it's going to be really exciting. Speaker 2: Nice. Yeah, that's going to be great. That's a whole new revenue stream. Speaker 1: Oh, yeah. Speaker 2: So something that you, I want to get into the idea of having a custom GPT, right? I think that's what most people know it as basically having a knowledge base for your business. But before we jump into that specific topic, something you mentioned earlier, you were talking about when you're prompting, when you personally are using ChatGPT that ChatGPT has context, right? You mentioned a context profile. So how exactly do you go about giving And LLM like a ChatGPT or like a Claude, how do you give them that context on you or your business? Cause I mean, you said, I think word for word, you were like, yeah, ChatGPT knows me and knows my business at a very granular level. Like how, how does it know all these things? Speaker 1: Yup. So, um, in, in the backend, this is called RAG. That's short for Retrieval Augmented Generation. And then there's also context windows. And then there's like, in those context windows, there's something called tokens that like you have a certain amount of tokens per thread that essentially, once you spend through those, it's not you're actually spending money, you're just spending context. It starts forgetting things from earlier on. And then that's when it starts hallucinating. So the reason why I say that is because when it comes to my personal profile, at the highest level, it knows me just because I've been using it. So like your storage, I mean, I think currently right now, I do have, like it says, my storage for my user is like, like out or whatever. So yeah, go in there and clean things that you don't want. So I got to do that occasionally. And I'll clean out things that like little junk chats that don't matter. And I just keep all the things that are important. And then you can also configure your user's actual context profile at a high level by going through the settings. And then you can add more context about your user and who you are and what you do. So I really went deep into that and I really just use, , I use ChatGPT to help build that out. In some cases, it recommends, especially when you're building custom GPTs and stuff, it wants the context profile like a JSON file. The structure of the data is really easy for the LLM to follow the path of the data and know what it's talking about, what aspect of what it's retrieving. So there's a two-part question there or a two-part answer. One, making sure that your history and your profile is clean and you only are storing the things you want it to remember. Going into your settings and making sure that all the settings and the questions is asking about your personal preferences and what you're using your profile for and anything you want it to have context on. You fill that out because that will always be there as like a high level context. And then at the custom GPT level, just really filling out those configurations. When you go to edit your custom GPT, use another thread that you can work with to kind of build a context profile. You can put in there, attach PDFs. I did find that PDFs are one of the best ways for the LLMs to pull the data. You know, anything you create via Word doc, you know, export as a PDF, make sure it's named nicely so it can be easily found, like just little things like that that engineers do when they structure code and things. You've got to kind of think like an engineer a little bit, like everything has to have a naming convention. Everything has to have a primer reason as you're loading this information into the custom GPT. All of that will improve your chances of getting the output that you're looking for out of it. Speaker 2: Right. And that makes a lot of sense because what I guess what I have and what I've done is I created actually used a context profile generator that was this guy on Twitter has some AI community and as part of his community, you get access to a I'm a context profile generator, which is literally, it's a custom GPT that just asks you questions about yourself and what your goals are. And then it creates, it spits out a JSON file based on that. So I told it, I was like, well, hey, I want to create a context profile for my writing, for my writing style, for my voice, for my, you know, the way that I write more or less. And so it asked me some questions. It's like, well, hey, What type of style do you typically use? What type of tone do you use? Give us some examples of your writing, right? And I just spent probably 25, 30 minutes literally just answering the questions that it asked me. And then it spit out a JSON file. That was pretty lengthy of just like really at the end of the day, all it said was, hey, here's some examples of Corey's writing. Here are some words that he likes to use. Here are some words that he avoids. He doesn't like to use emojis. Right. Just basically one big rule set for how I write. And so I took that JSON file and just saved it, like you said, into a custom GPT and also in Claude as in a project as the project file, because I personally think Claude is better at writing, but it's the same concept, right? So now anytime I want to If I want to write an email or write a blurb in my voice or in my tone, I can just go into that custom GPT and write it there. And it's so much better than it would be if I just went into like regular ChatGPT and said, hey, write me a blog post or write me an email, right? Because it's actually going to use my context, my tone, my style. Speaker 1: Yeah, more importantly like when you when you structure it the way that you did through that generator like it's really clear for the LLM to know what the rules are, what the guidelines are. When you just put a prompt in there, upload PDFs, it leaves it open to kind of make like make more reasoning around what to do and when to do it. But like what you call it a rule set and It really can kind of help it be more rigid and not hallucinate as much. But it will hallucinate occasionally, but like it won't hallucinate as much when you have that kind of a setup going. Speaker 2: Right. Yep. And so on the topic of custom GPTs, that's kind of the last topic I wanted us to touch on today. I know you've got extensive experience in building custom GPTs for your business and for your team. I just mentioned I've got one. The only one that I use is for my really for my writing style specifically. But I guess the question, Andrew, is What would somebody use a custom GPT for? What are some of the more common use cases that the everyday business owner would use that for? Speaker 1: Yep. The one that I see the most is. FAQ kind of bot that like, you know, you prompted a question and it pulls from your, from your approved like knowledge base and then it gives you answers. That's like the biggest one that everybody does. And then like, you know, after you have that built and you have that knowledge base, like it's really easy to take that and then put it into like a chat bot that you can like put on your website. Um, so that's like a really, that's a, that's a really common one. Um, I use it, uh, for marketing level things. So I have like a marketing GPT that knows All of YoDev's current strategies and, you know, where we're trying to go. And I do like, I tidy it because like, you know, as you know, marketing strategies change from time to time. So I clean it up and I make sure that it's like, it has what it needs to know about the brand and what we stand for, but then what our current initiatives and strategy is for like the current quarter. That can help me plan out like our YouTube content campaigns, our email campaigns, and it just kind of knows exactly what we're trying to do, which avatars we're trying to hit, and then can help us produce that content and even, you know, the plans and everything that my marketing team can then execute on. So I use it like that. And I see other people using it like that as well. A lot of our clients are using it. So they'll feed They'll feed tons of SEO research and data and all this stuff about their brand into the GPT and then they'll produce highly, really, really good, nicely written articles that are SEO ready and cut out, and not to cut out you SEO guys out there, but cut out the SEO discovery and guess phase. Get something really good out the door and then bring in the SEO vendor to scan after to then optimize further. Because really all you're doing in those first phases is they're doing the same exact thing. They're going to just use AI to produce a really nicely written article that meets your criterias for what you're trying to accomplish. And then a month later they're going to scan or two months later they're going to scan and optimize further. So you can cut out that whole noise in the beginning. And like have your content written and go out there and then use the tools to scan it later and optimize later. So those are, those are probably like, you know, since I do a lot of websites and marketing level projects, like those are kind of the, you know, the, the knowledge base kind of question and answer bot, GPT, the marketing strategist, and then like the SEO content writer kind of GPT. Like those are the ones that I see the most that people actually use on a day to day. Speaker 2: So really, you could take any vertical of your business, right, and say, you could even have, like, I've seen people have, like, an ICP, custom GPT, so, like, ideal client profile, where they've got, maybe they've got a document that just really spells out who their ideal client is. Hey, it's this industry, this revenue level, this many employees, like, just spells out to a T their ideal client. And then they kind of run all their marketing campaigns or run different ideas through that ICP custom GPT and then have the AI tell them, well, hey, yes, I think this would resonate with your ICP or, hey, you said that you want to target gem owners, but this ad looks like it would be better received by a spa owner, right? Or whatever. I'm just making up an example. But I mean, really, it sounds to me, correct me if I'm wrong, that It's almost like a custom GPT is most beneficial when it has a very, very vertical use case where it's not trying to do too many things at once. Is that fair to say? Speaker 1: That's exactly right. You do not want to give it too many objectives. You don't want to give it, like basically everything you're trying to ask it to do, don't leave it up to the GPT to make the final decision. You kind of have to have like, hey, Our output needs to meet these requirements. Boom, boom, boom, boom. The goal of this is to produce this when I input this, if that makes sense. You don't want to give it the option to just do whatever it wants to do. You don't want to say, hey, you're going to help me with my high-level strategy, my SEO content, my this and that. No, no, no. Each one needs to be split out as its own. Speaker 2: Right. Speaker 1: To be really effective. Otherwise, it's just going to get really convoluted and mixed up and it's going to get confused and it's going to really not be as valuable as it could be if you were to kind of specialize the GPTs. Just like specialized vendors, right? The person that offers everything, that promises the world all the time, typically isn't going to be the best bet to like really nail specific areas and get big impact, you know? Speaker 2: Right. Yeah, I think that's a great point because I think that's probably the biggest mistake people make with a custom GPT. They're just trying to do too much with one single instance of a GPT. No, that makes a ton of sense. Now, Andrew, I know you had, I think you were going to show us something. I think you were going to share your screen and kind of show a demo. Yeah, one of the ones that you built. So for those of you guys listening, if you're listening on Apple or Spotify or one of the podcast platforms, this would be a good time to switch over to YouTube. My YouTube channel at Corey Ganim and watch this on video because Andrew is about to share his screen and show us something pretty cool that he's built, a custom GPT that he's built. Yep. Speaker 1: Cool. So before I get into this, okay, so I will, I want to give you guys some context first. No pun, no like, you know, AI kind of thing. I'm going to give you guys some context. Um, so is this it? Yeah. Okay. Here we go. Unknown Speaker: So, oops. Speaker 1: Here we go. All right. So. Shout out to AI Meetup in Greensboro and the founder, John. He gave me access to this custom GPT that they built. The idea was students are cheating, or not cheating, but they're using AI the wrong way. Similar to what you posted on LinkedIn, Corey, they're using it to get the answers, but they're not actually learning why that's the answer. And they're also, you know, they're not really understanding the subject matter. So this guy is connected with a lot of universities and he was like, okay, well, we want to have our professors create custom GPTs and upload course material. The students can then go and quiz on the materials and then the professor can upload it time and time again. This is just a proof of concept. I'm going to get into something I actually built after this. I just want to show where it's coming from. For instance, this is AP precalc. Can you quiz me on unit one polynomials? And, you know, this is it's got its knowledge. It's going through and then it's like it's searching. Speaker 2: It's it's not like it's out there searching the Internet, searching all the information available. It's searching the the approved like. Right. Speaker 1: The approved subjects that and topics and like information that the professor, quote unquote, uploads. Speaker 2: Right. Speaker 1: Walks you through this quiz and you know, it's, it's cool, right? It's a, you know, I'm not that like, no, I love you, John, but I'm not that impressed with, with it. And it's because I'm an engineer. And that's why so, but like, it's useful. It's not scalable because the scaling model was every single professor, every single class needs to create a custom GPT and then we need to rely on the students to use their own personal open AI accounts to get access to the GPT. So no educational facility would ever bite on that. They would want their own system that they would allow you to log in with a .edu email or something. So now, fast forward to the event that I spoke at, which I will be posting on YouTube. So Corey, when that's up there, I'll share it with you so you can share with the audience. Speaker 2: Definitely. Speaker 1: So I took this proof of concept, built a front end web app UI. I connected a RAG backend, which we talked about earlier, and I'll talk about RAG at a high level just so you understand the demo. And then I replicated this but made it more of like an actual app, right? We call this the quiz generator. Speaker 2: And real quick for the people listening, if you're listening on podcasts, the demo that Andrew showed us that this guy John built is literally a custom GPT. It's an instance inside of ChatGPT where you can access it within ChatGPT. What he's showing us now that he built, As a way to essentially replicate and do better than that example is as an actual web app. So we're looking at a screen that's, it's not a ChatGPT interface. It's basically a website that is replicating those same functions. So sorry, I just wanted to explain that. That's good. Speaker 1: That's good context for the Spotify and like, you know, audio listeners. So at a high level, I just want to explain why, you know, one, I just talked about the scalability issues. That's one thing. But even if it was a small-scale operation, like why would we want to put it in a web app and connect a RAG backend? It's because, one, the data feeds live from the RAG backend. So as the professor is uploading course material or removing course material from a given topic, It's updating in real time. So that means that, you know, any student threads that are open in the app or anything else, like it's getting updates real time. So it can kind of like, it's working with the actual data. It limits the hallucinations because it's really rigid when it comes to like, hey, it's just using the data that the professors are uploading per class. And then it's scoped per class. So I'll show you how we're doing that. I'm tagging it like a topic. And then on the quiz side, it's not just static text that then you've got to copy into like a Google Doc and work with or whatever. It's like an interactive quiz that you take and then it tells you what you got wrong and what your percentage right is. So it's a more interactive experience versus the students having to like take it off of ChatGPT and like fill out their answers and work with it in their own tools. So first I just want to kind of show what we got going on. So I got this. This is like for in development purposes. So this is going to run a test and it's going to test to make sure that our back end is connected. Our Open AI API is working. And then it's also going to tell us how much documents are in the system currently. So I can run this. All right, so it says blob storage is connected. Our Open AI API is connected. All integrations are verified and successful. There's nine files in storage. So the first part of the demo would be the professor. So the professor needs to upload their information. Um, so if I come in here, where is it? There we go. You can't see this right now. Actually, Corey, I would kind of wanted to, uh, not put you on the spot, but you don't happen to have any, uh, eBooks or any PDFs floating around that, uh, I can upload in here. Speaker 2: I don't know if I haven't saved on my computer. Let me look. Speaker 1: I don't know if they're floating around online, like in your resource library. I know you have that. Speaker 2: Yeah, I've got a VA playbook, which is like a nine-page PDF e-book. Speaker 1: Do you have it right now? Speaker 2: I do, yeah. I can email it to you right now. Speaker 1: Yeah, email it. I want you to take the quiz on your own thing. Speaker 2: Yeah, let's do it. Speaker 1: Okay. Speaker 2: Let's see. All right, you should have it. Speaker 1: Okay, I'm waiting. I love tech demos. Speaker 2: Yeah, this will be cool because this is an ebook that I created back I think we made this one for the wholesale challenge. So this one is it's it's like it's called VA playbook. I mean basically it's a How many pages 15 16 page ebook on how to perfect hire and leverage a virtual assistant? Speaker 1: It's awesome. All right, so I'm gonna go ahead and Upload your VA playbook Okay, so that's up there and then we're gonna we got to give it a topic so Topic will be hiring the A's Actually, we'll do let me just do make it more general virtual assistants can't spell Here we go. All right, so it's going to upload an index. And to explain indexing to the audience while it's doing its thing, what a sophisticated RAG system will do is it will upload your PDF and then it puts it in what's called chunks. That is the actual technical term. It chunks your PDF into bite-sized snippets of data and then it tags it with a keyword or phrase that allows the LLM, like when you put in your prompt, it allows the LLM to quickly find it so that it's not wasting time going through the whole PDF every time. This is something that like, so you can control the chunk amount. You can, there's a lot of things, so it's like, How granular do you want it to go through? You don't want it to read through 20 pages every single time. When you type in, hey, I want to know about this, it takes your keywords and then it finds the chunks that only it needs and then it'll pull that information from there to generate the response. But this is what this means. The document's up there. It's indexed in the database and it's got 14 chunks out of that nine-page PDF. So now that that's in the system, I can go demo the student. So the student's going to take the quiz. So virtual assistants is an option here. So I can check the virtual assistant now and it's going to generate the quiz and now this is using OpenAI's API. It's using ChatGPT-0 or 4.0. So it's using that API to go and generate the quiz and then it spits it back to our side and it'll create the questions and then now we can go answer. Outside of a little UI flaw that I'm in the middle of fixing, I can check these off, but Corey, I'm going to run through this. Speaker 2: This is crazy. First of all, like just this whole concept that we just, I mean, just again, to illustrate for the folks who might be listening on audio, Andrew just took my, and actually I think it was like 14 or 15 page ebook about how to hire, manage, train virtual assistants. He uploaded it to this tool. Which took all of 10 seconds and then you click generate quiz. And now we've got a full, looks like four or five question quiz that is pulling directly from the ebook. Like these are all, like if I had to create a quiz based on this ebook, these are the exact questions I would ask. I hope I get them right. Speaker 1: Let's see, you're going to find out because we can get the score at the end. Speaker 2: So the first question, what is one effective way to ensure your virtual assistant feels ownership over their role? The options are provide them with a detailed job description, let them help create standard operating procedures, assign them tasks without explanation, or limit their communication with you. I'm going to go with number two, let them help create the SOPs. Which of the following is not a reason to consider hiring a virtual assistant? One, you have proven you can repeatedly source profitable products. Two, you are new to the business. Three, you have a demanding day job. And four, you want to focus on scaling your business. So which of the following is not a reason to consider hiring? You are new to the business. That's definitely not a reason. Question three, what should you do to ensure clear communication with the Filipino virtual assistant? Use complex jargon in your instructions. Provide vague task descriptions. Ensure instructions are easy to understand and or limit feedback on their work. I think it's gonna be number three, ensure instructions are easy to understand. Question number four, what is a recommended practice when posting a job ad for a virtual assistant? Option one, do not require any specific responses. Include a request for a specific word in the subject line. Ask for their complete work history up front or avoid specifying the skills required. That is going to be number two, include a request for a specific word in the subject line. And then question five, which is the final question, which of the following can help enhance a virtual assistant skills? Is it minimizing their training, providing opportunities for professional development, limiting their responsibilities, or avoiding feedback on their performance? And that's going to be number two, providing opportunities for professional development. I will say this, I thought it was going to be a little trickier. I thought it was going to try to trick me up. I guess it makes sense because it's pulling directly from, it's only pulling from the ebook. And obviously the ebook's only going to contain, I guess, best practices, if you will. So it probably doesn't have a lot to go on as far as like trying to trick you up with questions that aren't going to be. Speaker 1: One thing to think about too, so we got the same feedback when I demoed this at my event. That is a guy kind of easy to answer. That's because I haven't done any tooling on it. I haven't actually given, so it's only taken in data and generating a quiz. It's not like, I'm not giving it like, hey, this needs to be like college level, like answers need to be like, they're kind of be hard to distinguish type of a thing. So this is just like, this is very, very MVP right now. Speaker 2: Right. And think about, I mean, for, for the folks listening, right. Oh, did I get one wrong? Wait, what was, which one did I get wrong? Go down. So the question was, which of the following is not a reason to consider hiring a virtual assistant? You have proven you can repeatedly source profitable products. That is a reason though. That is a reason to prove it. Speaker 1: So there's a bug in the system. Speaker 2: I was like, wait, I was like, I know that's right. That was like the most obvious one. Speaker 1: So here's the rationale though, it's going again. Speaker 2: Yeah, let's hear it. Speaker 1: It's saying, if you have proven you can source profitable products repeatedly, it indicates you may not need a VA yet as you're capable of managing tests on your own. Speaker 2: Oh, interesting. I mean, I get where it's coming from, but that is actually the only point at which you do hire a VA. If you can do it on your own, that means you can train them effectively, whereas if you're brand new, You have no business hiring a VA. Speaker 1: That's where Corey professor goes, okay, it was pretty close, but now I got to go upload some more information. And then maybe like, I got to, I want to upload maybe a context profile. Like this accepts, I can make this accept JSON. It accepts TXT files now, so you can put your context profile with your rules and stuff and how you want the quiz responses to be for each class, topic, etc. That's where you can refine it a little bit more. There was an easy answer. It was pretty easy and then one was kind of like the AI just said, Noah, Corey, no, I'm right. This was a really fun project. It took me two hours to do, so it wasn't that hard. Speaker 2: Well, and think about the context too for people that are running a business where you're actively hiring employees. I mean, this is a really good way to onboard people, right? Is give them quizzes for their role, test their knowledge. I mean, yeah, it's not going to be the end-all be-all when it comes to training, but it is a good way to measure their knowledge and it's something that is trained specifically on your data. It's not like you're having them take some generic quiz on their role that's not specific to you at all. Speaker 1: We could integrate, for instance, instead of it manually being an upload, we could integrate with their knowledge base of choice, whether it be Confluence or Scribe or whatever else. With those tools and then use that and then scope it based on like folders or something like if it's a folder for a VA then we can just for the VA topic drop down it would source from your connected scribe that And it only hits the VA folder and all the SOPs that are in the VA folder, for instance. Speaker 2: And even think about having that as even just a simple chatbot, right? Because I can't tell you how many times my team will come to me with a question that I know for a fact is in our SOPs, it's just they might not want to take the time to go through the 30 or 40 SOPs that we have. But if we had a custom GPT that was trained on our SOPs, If they have a question about shipping, for example, they could just go ask the chat bot first, and eight times out of 10, they're probably going to be able to get the answer they need right there. So there's so many use cases for this stuff, and I think we've only scratched the surface, Andrew. I think we'll need to have you back in the future. Speaker 1: Yeah. But yeah, Corey, it's been a pleasure. I know we're kind of over time here, so I'll let you kind of, you can do whatever you do at the end here, but this was really fun. I had a blast. Speaker 2: Yeah, I enjoyed it too. I mean, Andrew, where can people, I'd say, where can they follow you? Because I know you're putting out a lot of content on this type of stuff. You're definitely more technically minded. So especially if people really want to get in the weeds, where can they go follow you? Speaker 1: Yeah, so the big place, I got two places, LinkedIn and YouTube. Just FYI, as you can see on my shirt, YoDev, Y-O-H-D-E-V. You search that anywhere, you're going to find me. You're going to find our YouTube channel, our LinkedIn page. That's where all the good stuff is coming. And a lot of the stuff that I'm getting into now is like with the rise of AI, there's also mass layoffs. So I started as a freelance developer and I scaled my business to, you know, we're breaking a million dollars. From zero to breaking a million, like I know what it takes. So a lot of these developers that are used to full time jobs now have to get into the freelance game. So a lot of my content now is also going to be around like how to run a freelance business, how to get your business going. I'm really good at sales and opportunity creation. That's the part that most tech people and dev people, they're good at the technical stuff, but when it comes to networking and making opportunities and then going to hunt and sell, That's where they lack. And a lot of our content is taken off in that area. So we're getting ready to train the next wave of the gig economy right now. I'm getting tons of devs that are laid off coming to me for assistance. So we're just going to start pumping out content around it. And then there's, of course, going to be business, AI and dev all kind of intertwined into one. Speaker 2: That's awesome. Yeah. And I think there's going to be so much opportunity for people that do have that technical background and that technical skill set. I mean, I think getting laid off from a dev job in 2025 is the biggest blessing in disguise that there could possibly be. Speaker 1: It really is. Because like, look what I can do in two hours. Speaker 2: Right. Speaker 1: Because of what I know, right? That is the advantage that engineers and devs have. They think that they're afraid right now because they think their jobs are being automated away. But you out in the wild have way more leverage. There's tons of companies, big and small, that would love to have a senior level developer engineer who's embracing AI to come in and like manage their tech stacks for them and you can charge decent Money to like be in there because I've done it and I've done it to the point where I've scaled a team that's doing it for me now. But it's very doable. And the time is now like this is this is the time this big opportunity and the window probably is about three to five years, really. So everybody's got to go get it right now. Speaker 2: Yep. And guys, again, that's YoDev on YouTube, Y-O-H-D-E-V and he is Andrew Hewitt on LinkedIn. We'll have links to both of those in the description if you're watching on YouTube and in the show notes if you're listening on Apple or Spotify. So, Andrew, thanks again for your time and I will see you on Tuesday. Speaker 1: Thanks, Corey.