#359: Blake Urmos on Building AI Tools for the Low-Voltage Industry

In this week’s episode of AU, guest Blake Urmos, Founder of Low Voltage Nation joins Ron to discuss How one simple ChatGPT project opened his eyes to the business potential of AI and led him to start building his own software tools.

This week's episode of Automation Unplugged features Blake Urmos, Founder of Low Voltage Nation. Blake first joined us back on episode 203 in February of 2022, and a lot has changed since then. Today, Low Voltage Nation has grown into a massive industry community with roughly 200,000 members, while also expanding into media, software, job boards, and now AI-powered business tools for the low-voltage industry.

In this episode, Blake and I discussed:

How one simple ChatGPT project opened his eyes to the business potential of AI and led him to start building his own software tools.
Why Claude Code, AI skills, agentic coding, and tools like MCP are helping non-traditional developers solve real business problems faster than ever before.
And how Blake built LVN Signals, an AI-powered platform that helps contractors uncover commercial low-voltage opportunities before they hit the bid boards.

Blake brings a unique mix of field experience, community building, software development, and hands-on AI experimentation. This conversation is a real-world look at what’s possible when integrators stop treating AI as a buzzword and start using it to solve practical business challenges.

Whether you’re already experimenting with AI or just trying to understand where to begin, you’re going to get a lot of value from this conversation.

So settle in and enjoy my conversation with Blake Urmos. Let’s get started!

Transcript

Ron:

Welcome to the Automation Unplugged podcast, the podcast for technology professionals featuring leading industry personalities. I'm your host, Ron Callis. Let's get started. Hello, hello there. Ron Callis with another episode of Automation Unplugged. Today is our very first inaugural AI Insights edition of Automation Unplugged. In this show, there's a brand new format. We're launching it today and we're going to be releasing a new show in this format once a month, every month. I believe we're going to generally do the last week of the month. And our goal is to try to bring you value and real examples of businesses and business owners and operators that have applied AI in unique, fun, interesting ways in their business to solve real problems. And so for today's show, our kickoff show, I could think of no one better than my good friend and industry veteran. We have Blake Urmos. He's the founder of Low Voltage Nation. He was last on Automation Unplugged a little while ago — he was last on episode 203, which we recorded back in February of 2022. And he's now here for our brand new AI Insight Show. So let's go ahead and bring in Blake. Let's get him introduced. Hello, sir. How are you?

Blake:

I'm well. How are you?

Ron:

I am good. We're both a few years older than we were, although you don't look like you've aged a day.

Blake:

I just ran ten miles, so I feel pretty good. I look refreshed. No wrinkles.

Ron:

Ten miles. That's pretty legit. My long mileage day is Saturdays. I run five or I run 10Ks, which is six miles, six point two. But ten miles — that scares me. Is that like a regular daily workout for you?

Blake:

Well, I drank a bunch of coffee this morning, so I wanted to get the jitters down for this. I haven't done a podcast in a long time, so I wanted to get the nerves down a little bit.

Ron:

Got it. Well, let's tell the audience who are you and what is Low Voltage Nation? Just a quick reminder to everyone.

Blake:

Yeah. So I started Low Voltage Nation back in 2018, I think. And, you know, so I was in the field. I was doing structured cabling. I was installing like nurse call systems, access control, and then I wanted to get onto a podcast. So I started my podcast, and then it just exploded into this big Facebook group. We have about two hundred thousand members in there, and then we do a bunch of media for other companies, for vendors, manufacturers, suppliers, but we also do software. So we've actually implemented a lot of AI stuff into our software. So that's kind of where the AI component kind of comes in. But yeah, I've been doing low voltage for over ten years now, but Low Voltage Nation for over six years.

Ron:

Amazing. Tell me about your AI journey. And I'm going to reflect — just for those that do go back and check on our previous conversation — that was kind of in the heat of crypto and NFTs and lots of other buzzwords that are... crypto's still a hobby of mine, but it's a little less in fashion these days. But AI is very much in fashion. It's changing the world — it feels like it's changing at light speed. What has your journey been with AI?

Blake:

Yeah. So I guess the last time we talked, ChatGPT wasn't even a thing. And we started doing some NFT stuff, but we never really followed through with it, which was a good thing.

Ron:

Amen.

Blake:

We had a twinkle of a good idea and we said, let's simmer on that idea. And then we decided to not act. And I think this AI thing — I think this is it. This is the big one, isn't it?

Ron:

It's here to stay. And it's funny. Feels like maybe the biggest thing ever, to be completely honest.

Blake:

It's a big deal. I remember a few years ago I was on YouTube, I was trying to find stuff about AI and you couldn't find anything. And this is like five years ago. I just wanted to learn about it. And you couldn't really find anything on it. But then once ChatGPT came out, I started typing in — I was like, hey, I need some JavaScript. I was doing this job board for Low Voltage Nation. Some client of mine, they wanted to put a banner at the top. I was like, hey, I need some JavaScript to put this banner. And it just spit out the JavaScript. I'm like, what in the world is this? And then it was like a four hundred dollar project — a quick little thing, it took me like ten minutes to do. Like, I didn't know JavaScript. I couldn't figure any of this stuff out, but ChatGPT just spit it out and made me four hundred dollars immediately. Like, I was hooked. As soon as I saw that it could write code, I was like, what else can I do? So I started building my own job board because I was paying, I think it was, like, five hundred bucks a month for this job board software. So I just rebuilt it. I just went into ChatGPT and said, hey, I need to learn how to make a job board. And it just started spitting out all this Python scripts. I was like, what? Like, where do I put this? So I had to keep asking, like, where do I put this? It said, put it in Railway, put it on Vercel, put it on these places. I was like, okay. So I just followed the instructions and hours and hours would pass. Like, this stuff is so addictive. Like, once you see it spit out the results and it works, it is like the best dopamine hit. It's like the challenge plus the dopamine, and then you get actual results and money from this whole thing. So ever since then — and then once all these agentic coding tools came out, like Claude Code, Codex — it just does it for you. You can just type in what you want it to do and it creates entire back end and front end infrastructure. That was kind of my journey in a nutshell up until now, but it was that first JavaScript moment with ChatGPT. It had me hooked immediately.

Ron:

To the best of your ability — and you and I are both laymen here, so we're not AI experts, although we both work with it and dabble. Our team here at One Firefly certainly is doing some AI things. But assume people listening may not even understand the immediate difference where you said Claude and Claude Code, ChatGPT and Codex. What is the difference between Claude and Claude Code?

Blake:

Okay. So this is actually really confusing because Anthropic, the company that makes Claude, they have their chatbots. You can go to claude.ai. You can just do your typical ChatGPT-type chatbot kind of things, and it'll — you can put in pictures and it'll spit out all sorts of information. It'll search the web. So that's like the front end chatbot. But with Claude Code, you're in the terminal. So if you're terminal native like me — I have a big Linux background, and I know some programming languages — if you know the terminal, you can get in there, and it'll just loop through a problem. It has file system access. It can run commands. It can do anything on a computer, and it just loops through and reasons through it. Whereas just regular Claude AI or ChatGPT, it's just like a chatbot. This is completely different because it's called agentic coding, and it just has access to tools, reasoning, and it can connect to MCP servers or APIs — it can connect to anything, basically.

Ron:

So let's — I'm going to totally put you on the spot. It's not a gotcha question, but it's like a brainstorm. Let's assume people watching or listening to us are business owners or operators within low voltage businesses, security businesses, electrical contracting businesses — all sorts of shapes and sizes out there tune in. What would be applications — maybe low hanging fruit applications — to where they might use Claude in their day to day? And where might they want to deploy Claude Code to solve a problem? What's a simple example?

Blake:

So I can do a screen share here in a second. But if we want the most basic — let's say it's an integration company that does low voltage. They want to do invoices and a quote. Okay. So they could use a Claude skill. So a skill is basically just a markdown file. It's just a text file. Do you know what a markdown file is? It's just a very basic syntax where you can have bold characters. It's like HTML, but even simpler. It's an easy format for AI to read. It's preferred over PDFs or other file formats. It is like the programming language for AI, because it's basically just like a prompt that has a lot more structure than just typing, "hey, make me a website." This is like actually: here's all the queries, here's all the structure, here's everything that you need to follow to execute that task. So Claude created a standard — it's kind of a standard. A skills.md file. And other AIs use it — it's like an agent.md file. So they have these kind of different naming conventions for it, but skills.md is kind of what everybody landed on. So Anthropic created this standard where you have this very structured instruction — it's like a big mega prompt. And let me share my screen to kind of show you what I'm talking about.

Ron:

Let's do it. We'll see if technology behaves.

Ron:

Folks, we're living dangerously here on Automation Unplugged. We normally aren't sharing screens, so let's give it a shot.

Blake:

Yeah, this might go poorly, but let's try to explain this. So for example, I have this skill called Ad City. And this is something that I continue to work on almost every day and improve. But what this does is it goes out to a city like Detroit, Michigan, and it will look for all the permit data and APIs — whether it's a government website or maybe looking for an RSS feed for anything related to low voltage. And you can kind of see — I'll show you what it looks like. This is the markdown version of it. But if you preview it, it's a little more human readable. But it is this massive document that tells where to go find the data, and this is a repeatable process. So if I were to run Claude Code — you can see that I'm in the command prompt right here. So I'll run this Ad City skill. And what it's going to do, it's going to go start at the top, it'll look at everything that it needs to access, whether it's a database, whether it's my internal documents to see what cities are already added to my system. So it shows — I gotta be careful not to expose anything.

Ron:

Yeah. Yeah. We can add blurred areas to the video in post production here if we need to.

Blake:

Okay. So here's an example. San Bernardino. Yeah. Okay. So this is not in my database. So I'll just type one, and it'll know to run through this entire skill right here, and it will test every single endpoint, any API endpoint. So right now, it's looking for this ArcGIS. That's a pretty common permit kind of framework, kind of platform for government sites. So it'll go out, it'll see if there's a publicly available API, and it'll add this to my database. It'll add this to my scraper. So every day at three AM, it'll run through this particular city, and it'll add permits to my city. I probably shouldn't be telling people this because this is — I generate revenue with this. This is what makes me money. Because I have over three hundred scrapers, and I add one every single day. So every single day — it's called a cron job — so a cron job is like a scheduled task. Every single morning, we add a new city to our database. So now we have about fifty thousand — these are all low voltage projects. So it's not only looking for any permit. It could be like a random — you know, put up a light pole or something. But it'll search and find anything low voltage related. So if it's a fire alarm, access control, fiber optics — this skill right here is what is instructing Claude Code to do that. So does that kind of make sense? I was pretty rough, but.

Ron:

Yeah, so I'm going to restate it, maybe ask for clarification. So you're in Claude Code. And you've actually created a product. And that product is — you're selling permit data in cities, and so people can come to your Facebook page. Is that how they would subscribe? Talk the business side real quick. If people go, this is really cool — how would they get access to this or subscribe to this or sign up to this?

Blake:

Yeah, I didn't want to make this a sales pitch, but since we're on it, let's go ahead and do it.

Ron:

I went there. You didn't go there. I'm pitching for you here. And this, by the way, this completely wasn't even planned. It kind of was like ten minutes before the podcast.

Blake:

Yeah. Ron put me on the spot. He's like, hey, share your screen — I see you doing all sorts of cool stuff on LinkedIn. Come on the podcast and show us something cool. Yeah. So this is LVN Signals. So lowvoltagenation.com. This is our new product, and this kind of goes back to what I was talking about with creating the job board. So I started with creating the job board, and then I refactored the whole website. We added projects. One of the things that I saw in our Facebook group is where people are adding projects and they're adding jobs. Those are the two biggest problems we have in our industry, at least from the Facebook group side — people want jobs. They want a 1099 or a W-2 job, and then all the companies want projects. They want to get — so it's like a double sided marketplace. So we basically started with the job board, and then... [Screen share] So here's our job board, and then Signal is where all the permits come in. So here's an example right here. This hospital project — where is this located? This is in Miami, Florida. So what that skill is doing is it's figuring out, like I said, the low voltage side of it, and then it runs through another LLM. So another large language model. I have it running through another Claude. It's called Claude Haiku. It's a kind of a lower cost, large language model — less expensive model for tokens to process the data.

Ron:

Yeah.

Blake:

Because right now, this is running through Opus 4.6. This skill right here. So it has the best, highest end of the Claude models.

Ron:

Correct? So that's probably the most expensive at the moment?

Blake:

Yeah. So I pay the two hundred bucks a month for Claude Max. And that gets you — I never run out of tokens, basically. Even with having these run every single day, and I'm coding all day long too. Two hundred dollars gets you quite a bit of tokens. So I'm running Opus 4.6, and then there's Sonnet 4.6, that's the one below it. But then for doing kind of the grunt work of just figuring out — okay, this is a hospital construction project, there's low voltage, there's nurse call, there's all this stuff — so it'll score it and be like, okay, this is an estimated value of one point five million dollars, score ten out of ten, just an arbitrary scale of, okay, is this like a low, kind of weird, rank one, or is this like a big project worth millions of dollars? That's a ten. So it runs through that whole scoring process, and then it'll put it in here. So here's Low Voltage Nation Signal. We've got forty-two thousand projects so far, and then it adds hundreds every single day as we add more scrapers.

Ron:

It's auto feeding into your website.

Blake:

That's correct. Yep.

Ron:

That's just through... magical Blake programming. How — at a simple level — would you describe... you have a Claude Code skill executing workflows daily, then being piped into an API, a separate API with Claude Haiku cleaning up that crawled data. And then how are you piping it into the website at a high level?

Blake:

So it knows to go to our database. So this skill — all of this, what you're looking at right here, this is an IDE, a developer environment. This is — Antigravity is Google's IDE. It's based off VS Code. It's kind of a standard developer environment. We're using Antigravity here at One Firefly. My son's doing software development in Antigravity. Folks, if you're listening, you can do it too. It's not that scary.

Ron:

Yeah. Antigravity looks scary. It looks very scary.

Blake:

Yeah. For sure. And that's a good and a bad thing. It's good for us because we kind of have a moat, for now at least. But yeah, Antigravity is free. It's based off of a free open source platform called VS Code. And you don't need this for Claude Code. You can run it in any terminal. That's what makes it so amazing is because you can run it on Windows, Linux, macOS, just in the terminal, and it has access to everything. So to answer your question, it knows within this skill to pump it into the rows in my database. It's using MCP — model context protocol. That's an Anthropic standard. You can use regular API calls and just push it to — I use Supabase, which is a PostgreSQL database. So you can pump it to that, however you want to do it. If it's on a command line, and if it's a tool on your computer, Claude Code can use it.

Ron:

And then from that database, how are you getting it pulled into the website and looking attractive and happening automatically?

Blake:

So I use Vercel. Vercel is just a serverless platform for Next.js. So Next.js is a JavaScript framework — it's based on React. So it's just a hosting platform, basically. So my Vercel is connected to Supabase, and that's it, really.

Ron:

Does that make sense?

Blake:

Yeah, it does.

Ron:

So at a high level for the audience that goes, all right, I'm not a programmer — how would you at a high level describe what you've done? Let's step above the code line. What was the problem? How long did it take you to solve that problem? And what's been the result like for you and your business and your client/customer base that's benefiting from this?

Blake:

So the last job that I had when I was in the field, I was actually doing sales. I was a sales director for a company in Nashville. And one of the biggest problems was getting projects. So we had ConstructConnect. There's another one that we had. There's like two platforms they paid a bunch of money for, and it didn't get the jobs that were really easy to bid on because they already hit the bid boards, and then everybody already knew about them. So that was a huge issue. So I figured — go straight to the permits, and then figure out who the GC is, figure out who any of the contractors, anybody, the private equity firm, like whoever is involved, and start building the relationships with them so you know before they hit the bid board. So that was like the big problem. Like, there were all these platforms out there. They're my competitors now, basically, but I wanted something where I had early signal. That's why I call it Signal. So that's the problem. And then I solved it with code, with Claude Code, basically.

Ron:

What's been the business impact for you at a high level and for your clients? How do you measure that? Or how do you measure success or ongoing success?

Blake:

Yeah, so we released it just a couple months ago. I've done some on-site visits, and what we've done is I get them three projects. Whenever a customer signs up, I get them three projects immediately. I'm like, hey — I ask them four questions. Like, what's your service area? What type of projects are you doing? What is the estimated value of each project? And then I try to get them like a big win initially. I'm like, here's three projects — go call these people and go get them. That'll pay for like three years worth of the product. If I can do a white glove service, do this for every single customer that comes in — I send them the email, we go back and forth, we hop on a call, maybe I do an on-site visit — but I want to get them a big win initially so it pays for the product within just a month.

Ron:

That's amazing. Is that database delivering mostly commercial projects? Is there any residential there? What's kind of the project types that are being pulled into the database?

Blake:

Yeah, it's all commercial. The only residential stuff is — okay. So we used to have all the community gigs, as we called them. So anything on the Facebook group, I actually had a virtual assistant go in, copy and paste from the Facebook group. So we actually have three thousand projects in the system, but I hid all of those. Because here's the thing — when somebody would sign up for Signal, it would show these Facebook group projects and it would link out to them. So then they would have to go to Facebook and try to deal with the person there. And those are a lot of really small projects. Like, go hang a camera at this one person's house. It's like really small stuff, but some commercial stuff. But I got rid of all those because they were kind of just low quality. And actually a couple customers, they churned because of it. I was like, oh, that's not good. So I stuck with the commercial stuff.

Ron:

Got it. Makes sense. What's — maybe we'll close out here with two things. What's a tip or recommendation for folks out there maybe at the earlier stages of use of AI in their business? And what would be a tip or a piece of advice you'd give them? And then we'll close out with your contact details so we can make sure everyone knows how to reach you.

Blake:

Yeah. Claude Code is the best skill you could possibly learn. I'm convinced. It's very scary because it's the command line, but if you can figure out how to at least create a skill in Claude, you can just use the chatbot. Skills are actually in that chatbot in Claude AI. I don't use ChatGPT anymore. I know everybody's like, "oh, chat..." — it's like the Kleenex.

Ron:

Yeah, exactly. It's like the generalized trademark. Yeah.

Blake:

But Claude is so much better. They have completely dominated the market on the coding side, and coding is what drives all of this. Once coding is solved — and it almost is at this point — they're starting to recursively self-improve. So if you can learn coding and Claude Code and just Claude skills, that's three things. Learn all three of those. Yeah. Those are the most valuable skills I can think of right now. Or go into the trades, because those won't get automated as quickly.

Ron:

There you go. I love it. Blake, thanks for coming on Automation Unplugged and our AI Insights feature. How can we get people to you? What would be your recommended paths? And I'll put the contact info on the screen as you say it.

Blake:

Yeah, lowvoltagenation.com is the best place. And then everywhere — Blake Urmos, all lowercase, B-L-A-K-E-U-R-M-O-S. I'm on LinkedIn, X. X is a big one. I know a lot of people don't like X, but that's where all the AI stuff is. That's my second tip — get on X. If you're serious about AI, get on X.

Ron:

I'm actually — with Allison's help — I've been a lurker on X for years and years because of some of our mutual hobbies. I'm actually myself going to start posting on X imminently here. Probably by the time we launch this show, you can go find me — Ron Callis. I'll be on there as well.

Blake:

Yeah, you don't have to post on there. Like when I post, I get three likes, if that, on a good day. Just observing because everybody's on there. All the news is there. All the breaking news everywhere in the world. LinkedIn is always two weeks behind and it's just like, I don't get any value from LinkedIn. I'm sorry, but I don't. But everybody's on there. Even like, Sam Altman — like, he hates Elon. They hate each other. But Sam's always on there posting his OpenAI stuff. So get on X.

Ron:

Get on X. And what about a phone number?

Blake:

Phone number for me? Yeah. Are we going to share a phone? I have a phone number here. Should I — should I not share that? Oh, I'll keep that.

Ron:

No, don't call Blake — email him or get on LinkedIn or X. Get on LinkedIn or X. All right. I've got the Instagram floating here. We'll make sure we drop those other handles down in the comments. Blake, thank you, my man. You're an awesome sport. We told you as we were jumping on here, we prepared folks a little bit, but we also winged it a lot. And we both acknowledge there's a lot of opportunity to help our industry become more aware of what's happening out there and what's possible. That's my goal with this new feature. Blake, you're an awesome sport. You said yeah, man, let's go. And thanks so much for joining me on the show, man. I really appreciate it.

Blake:

Yeah, I was going to ask you like one hundred questions because I wanted to know what's going on in the industry from your perspective, but maybe some other time. But I think that's more interesting because you've got so much insight because you're actually doing it. You're doing the thing. Like, you're the guy now.

Ron:

Well, we'll make — how about this? We'll close out this show and we'll book another one. You can ask me questions and we'll make that a part two.

Blake:

I think we should do it.

Ron:

All right, let's do it. All right, buddy. I'm gonna close out now and we'll get that scheduled.

Ron Callis is the CEO of One Firefly, LLC, a digital marketing agency based out of South Florida and creator of Automation Unplugged. Founded in 2007, One Firefly has quickly became the leading marketing firm specializing in the integrated technology and security space. The One Firefly team work hard to create innovative solutions to help Integrators boost their online presence, such as the elite website solution, Mercury Pro.