This AI found a leak in a bill nobody opens

Nicholas, it's great to meet you. great to be on. I look forward to our conversation.

Yeah, long way, long story. So I've spent most of my career in technology. So I I've been I started off my career as a software developer and then over time shifted over to product management. And I've been doing product management for about twenty, twenty-five years. During that time, I wanted to get into a side gig. So I invested in these gyms or studios called club lottties and they started doing really well. And after they, you know, it was about six years into it, I decided, hey, maybe I need to focus on it full time. So I quit my technology job and I started doing Club Lottis, which then I got a good offer for them. So I sold them at a a nice profit. And then I had to find something else to do. so I started off finding people from around the world and helping US businesses hire them so that they get really talented people at a low cost compared to the people we find in the United States. So that was the start of our business. We were a recruiting company, we still are, and we hire both in the US, right, for in-person. folks. So we have a we do a lot of healthcare recruiting. We've hired a lot of people in healthcare in the US, nurses, care coordinators. And then we also hire people from around the world and place them in the US. Now during this time, AI came about in the last year, year and a half is when it got really popular. And we realized that the number of people our customers need Is going to go down because AI is going to replace some of that talent. And so we said, hey, why not us provide some of that AI? So I went back to my software development roots and started building AI automation for folks as well. We first built it for ourselves and then we're now building it for other people. So that's my story of where we are in at Seamless Assist.

Yeah, great question. So our historically our customers have been gyms. And we so because I was a gym owner myself, I have a lot of contacts within the gym community. And so I went first went to them and you know tried to find use cases for them to build. So a couple of really cool things we've built there are an an AI dashboard. With all the GIMP's numbers. So it's if you take any business, any business has KPIs. And in order to understand those KPIs, you need them all to be displayed to you in a nice manner, an easy to consume manner. And AI is a great way of creating that mechanism. On top, so that creating KPIs in a dashboard was the easy thing. You can tell Claude a give Claude a command now. And build a rudimentary dashboard within one prompt, right? But ours are more sophisticated. We build entire database layers, pull data in, then ingest that on a regular basis, and then display it. So it's a more sophisticated thing than just a cloud prompt. But on top of that, we built this AI question type system where you can Prompt the dashboard like you would any chatbot, right? So like you would ChatGPT or Claude, you can type into our dashboard and say, hey, how was my studio yesterday? How was my trial class yesterday? Or what are my projected sales over the coming month or the coming year? And the AI then uses all your data and goes and fire formulates an answer for you to go and make sure you have the right next step for you. Right. So it's a a thing that we built on top of just displaying what your current data is. So that's one one use case for franchise consultants. We're trying to figure out you know how to create match, if you will. So people looking for businesses and franchise consultants, they typically are left to hey, the franchise consultant thinking, hey, I know this guy and he could use this business as a as a great stepping stone. Instead of a person doing it now, AI can quickly compare the candidate to 850 businesses and and and make and create a match. So that's another example of of a use case. We're building some for real estate. that's also a thing for me. I've built a property management software that helps collect all your information from all your properties, take payments, those kinds of things. So our c our customers are varied. we build custom automation. We're trying to get into products now so that it there's more recurring revenue. and so that's that's where we are in terms of overall AI automation.

Well, the one thing we've currently launched, semi launched, is a very simple product for ally personal training. It's a a set of gyms here in the US. And what we're doing for them is taking data from one spot and moving it to another spot. As simple as that might sound, that entire system Currently has people like very expensive directors come in every morning and hand copy that data over. Right? Now there was another existing system that didn't quite work for them. so this became a simple but easy thing for us to go out and and solve. it's got a a limited set of owners, but it also is you know, recurring revenue, software, good margins. so that's been encouraging me. I and trying to figure out what my next thing is on top of that. So that's a a recent announcement. In fact, I'm going to a conference next week to go launch it.

Yeah, how do they get started? Yeah. So I think you know, a lot of you'll hear a lot of the standard, you know, set up your cloud.md file, teach the AI about you, have it ask you questions and create that mechanism. More advanced people talk about loop engineering now, you know, graph engineering. Harness engineering also, right? And I've I've built all that, right? I think to get started, simply just go out and try things. What really helps me learn is go out and build the things, and the time you spend will actually help you learn more. So there are certainly go on YouTube, you'll find free lessons, right? Pick Any one of them just to get you started. And then pick your thing that you do in the in your day. Like a very simple thing for everybody to do is get a daily brief. Right? A daily brief of all the stuff that's going on in your email. If you use like WhatsApp, you know, have it read that. and it'll create a thing, a small little thing for you. It also thinks, right? AI is great about reading that material for you and making a decision on it. So as an a very useful example, a couple of weeks ago there was I have a couple of rental properties and there was extra water usage in one of my rental properties. Now last year I learned because of leaks in my other properties that extra water water usage amounted to leaks. This time AI f saw that extra water usage and noted it right in my daily brief saying, Hey, there is a possible leak in your property here. Go go check it out. Right? Things like that are are insane, right? How does

And how do you you know I would never have thought about it. I would not have read that email, frankly. Because email automatically gets, you know, I get so many emails that it just it was a utility bill. And I don't read because they're all on auto pay, I don't read utility bills. But AI went and looked at it and highlighted it for me. That kind of stuff had not happened ever for me before. So just just an example. So I think. Simply start doing it, start prompting it, and then build yourself a brief and that'll see you'll see in your daily life how it begins to change your life.

That's right. That's right. That's right. Now the more re the screenshots are great. The more recent things I've found is that both Cloud and ChatGPT will go read your browser by themselves, right? So you don't have to go paste screenshots anymore. It'll go what depending on what you're building, it'll go read the thing yourself.

Yeah, yeah. So that's one of my proud pieces overall, right, in my set of agents. It started off very simply. My marketing guy s used to spend a lot of time on blogs. And I always wondered I mean, you go he was doing the right thing. He's great at what he does. He used to find a topic, use AI to write it, and then humanize it. Make sure he posts the right links, backlinks in it, the right images, and then go onto the our WordPress website and then post it. And all those other steps around him just took so much longer, right? I said, Hey, we can have AI do all those steps. Right. So that's the first thing we built. We built a blog writer. That would first simply just write a blog, which is simple. You know, you can with the prompt you can write a blog today. But then our prompt was sophisticated enough that it put all the right backlinks in there and it created a full-on blog post. The next step was then, hey, how do I post it to the web automatically? So we went step by step. We wanted to keep an a human approval, like a human in the loop always, right? To say, hey. I want to make sure I know I trust my AI, but I also want to make sure it's saying and doing the right things. So we kept that for a while, right? So block created, human approves, then we built the the posting onto the web automatically. So now the blog piece itself writes everything. And our our guy Navniche reads it and then he hits post. It automatically posts onto the web, right? Approve or post. Then we went deeper into it. So we said, hey, let's create monthly, we'll create a marketing plan. So we first started off with a small set of posts. Let's say we'll do 14 posts in a month first. Right. So what the system used to do, it you used to plan for the month these 14 posts. It would put it on a calendar to say, hey, these are the posts. So Instagram, Facebook, LinkedIn, and blog, right? All of them. And then Once it went there, the human in the loop, in my case, Navnish Magai, he would go look at the post, approve it, and it would post itself. Then we found a problem with hey, Instagram, Facebook need images. So what we do for him is the AI generates a prompt to generate an image. So he then takes that prompt. And then manually generates the image elsewhere and then puts it back in and then it hits post. So once we went, we went one by one, so LinkedIn, Instagram, Facebook, right? And then we said, hey, let's go full on. Then we went up to the 30. Now we are at hey the plan generates all the 30 posts at once. And that plan is based off of a strategy for us overall, right? Hey, as a company, we want to focus on healthcare recruiting this month. And some AI automation this month. So the plan will divide up the number of posts and create a calendar which says, go review these on a daily basis. Navniesh, the man, human in the loop, right? He goes and approves each of those posts. There is still we we kept, we've so far intentionally kept parts of it manual so that we approve it. So I approve the overall plan, right? and then it goes to well he first approves it then I approve it and then the system creates the entire calendar and then he gets to approve each of the individual pieces before they go out. So that it takes out because the posting is automatic, he doesn't have to go in and type a bunch of things to actually post the captions generated. The only thing he does still manually is because he's a perfectionist, he when he puts the image in, he goes And creates a nice image. But he has a starting point for an image. There's already a prompt, a very detailed prompt built out for him to go generate the image. So yeah, that's that's how we went to 30 posts in a month. But try we're still always fine-tuning it, of course. So what when we change our overall strategy, it has to change all the posts that go along with it, right? So that plan is what triggers the rest of the activity for the month. And then the human in the loop make it just easier for him to go hit approve and then it goes and posts automatically.

That's right. That's right. It does, it does.

Yeah. You know, two things. That of course is something that we all have to watch out for, right? So as soon as the context gets above a certain point, I've gotten into this mode where I try to refresh the context. And in order to do that effectively, I have created a a project memory type second brain. So it Essentially I start off by saying, Hey, go read all this from project memory and we're gonna start a fresh context. And it reads that and knows what it's been working on all this time. And so it all comes back to it. And then as I work through a a particular chat, I use two AIs by the way to talk to each other in order to to build something. Right. So there's a review always happening and both are reading from the same project memory. So I have Cloud code in one window. I have Cloud Chat in one window. Sometimes I have ChatGPT and Claude Codex and ChatGPT, depending on what it is, right? Talking to each other there. so both of them read from the same project memory and write back to the same project memory. So that way it's always built in and As context gets full, I do refresh the chat very regularly. I try to go once a day. So I'll go start a fresh chat every day. So I date my chats now so that I know I'm I'm refreshing context on a regular basis. I'll have a date up front and then the topic of the the chat as well in the name of the chat so I know what I'm working on. Because I'm usually working on two or three things at the same time. The other thing that I've seen it not always do well, or people not do as well as they should, right? Is look for all the edge cases, look for all the the error conditions. a lot of folks I've seen, at least out in this in social media, they don't always go and say, Hey, what happens when we go to this edge and what happens when we go to this edge? Cloud code itself does that, a lot of it, right? And and that's great, but you as a human need to tell it what your conditions are and have it think about it before you go off and and go out and build it. So yeah. Memory context, great one. And then error cases, edge conditions, you know, think about those deeply so that you build those. starting out from the spec itself. So the the the a learning that I've gotten over all these times of trying out and making those mistakes is I used to be a product manager. And as a product manager we write good specs to give to our human software engineers. So I started doing that. I started writing really good specs to give to Claude So that it knows what the end result should look like. And then when it starts designing, I actually have it not write any code at all until it builds a design that I can approve, like a technology design, right? And then once that is done, then I'll have it go start writing code. So it takes me a few days. Sometimes it'll take me a few days just to get the spec done and the design done before we go off and start writing a single line.

Yeah. only on LinkedIn, mostly on LinkedIn. so you can find me Sheetal Guti Goli LinkedIn. I'm the only one there. and then seamlessassist.com as the other place where you can see our stuff. We do have the company has other social media, but I'm not on there, so but people can reach out to me directly on LinkedIn.

Thank you. It was great talking to you, Micholas.