Introduction

Hello everybody, my name is Rob. Today I'm going to talk to you about how I accidentally built a business from a UI wrapper.

For those who don't know, a UI wrapper in the dev world is you have some software and then you make a UI to make that software pretty, and that's what I did.

I'm not a smart guy, I don't know how to make LLMs, I don't know how to train models.

What I do know is how to take hard work that other people have done, make it pretty, and then package it it up.

So, the app that I made is called Voice Creator Pro. I'll talk about it in a sec.

I've been focused on AI tools. I'll talk about a few of them that I've made.

Basically, for the past two and a half years, I've gone really deep into AI and made a ton of tools.

None of them have been particularly successful, except for this one.

Ex -engineer, now vibe coder.

I haven't really written a line of code in about two and a half years. Everything Everything has been completely AI -driven, and I don't think that's going to change.

Background: From Engineer to AI-Driven Builder

Early Experiments and Side Projects

So these are some of the things I've worked on before.

This is InstaRiz. One of my buddies got divorced and he found himself on Bumble.

It used to be pretty common that women would use filters on their photos on dating. Men didn't really do that, so I thought, why don't I make an app that lets you take a picture of yourself, change the background to make it look like maybe you're in Egypt or maybe you're in Paris or maybe you're just wearing a really nice suit, and then you can use it

on your dating profile.

Shaper, this is a Chrome extension. As a web developer in the past, many times I would go on websites and they would be broken or annoying, dark mode wouldn't work very well. So I thought, hmm, why don't I make a Chrome extension that lets me modify websites?

So if I go to a website and I don't like it, maybe I want it to be in dark mode, I can just tell the AI, make this website dark mode, and it'll do it.

It turned into a more advanced automation tool, so it can fill forms for you whenever Whenever I'm issuing like a DCMA takedown, I just have a button, fills out a form, and all of it happens with AI.

And then I also made this assistant that runs on your watch. It's called Hopper.

The idea was that I could leave my phone at home and just take my smartwatch with me and have an AI that works independent of my phone and can do all the things that agents can do.

But none of those found much success, except for Voice Creator.

The Breakout App: Voice Creator Pro

This is an app that basically lets you clone your voice, there's a few other things, transcription, but the main functionality is you record your voice and then we can create a clone of it and all you need is four seconds of audio.

It works on your device, it doesn't use any kind of subscription or cloud APIs, it runs completely on your own machine. One time purchase, no subscriptions as I mentioned.

Supports 600 languages and it can do voice translation, so if your native language is Spanish, but you want to talk in English, you can clone your voice in Spanish and make the AI talk in English. You can also do audio books, dubbing, there's a whole host of things that you can do when

you have voice cloning.

Core Features and Positioning

Why a UI Wrapper Became a Business

So why did I make this?

The story is actually, this is what the interface looks like if you find an open source voice cloning model and try to run it yourself. yourself. It's not the best looking.

And the more complicated you get with these open source models, the worse the interfaces get. So this one behind me now is for music creation.

And then the other reason is some people really don't like setting up code and running Python and doing everything on the computer.

This guy, Calvin Tor, he just wants an effing EXE. He wants to click one file and have all the hard work done for him. So I thought I could do that.

Making Open Source Voice Cloning Usable

Signals It Was Working

Now, given that I've worked on a lot of projects, and none of them were particularly successful, a few things stood out about this one that made me realize that, hey, maybe I'm on to something.

Signal #1: Unexpected Early Sales

The first one was sales. Within 24 hours, I got the first purchase, and so it's a desktop app. It runs on Windows and Mac OS.

This is a picture of the Windows store where the app is currently hosted, and I don't know if you can see, but most of the apps there are free, and there's one app that's a dollar, And then in the top right is Voice Creator for $77.

If you would have shown me this before I built Voice Creator, I would have never built it. Because I would have thought, who is going to spend this much money on an application on the Windows Store? Nobody ever hears about people spending money on the Windows Store.

So I guess if there's a takeaway from today, build an app on the Windows Store because there's a lot of people who are willing to pay money if the app at least looks and functions relatively well. well.

Signal #2: Positive User Feedback

The other success signal for me was feedback. Very quickly after launching the app, I got nice emails from people. People saying, wow, congratulations, this is a very useful app, or hey, can you add Italian? Can you add Hindi? I'd love if you had these languages in your cloning app, I'd buy it.

And this is relevant because for the app I mentioned before, so Shaper, this is the Chrome extension that I made. Vibe code websites, right? Access AI in a sidebar. This is a very heavy, very obviously AI -based tool.

And the only feedback I ever got for this in the entire year of running it was, fuck AI, what the fuck? And the guy didn't even leave an email, so I couldn't even respond to him to ask him, sir, what do you mean?

Signal #3: Piracy as Demand Validation

And then the final success signal was piracy. So somehow my app got about pirated. It ended up on a popular pirating website. In between Adobe Photoshop and Avast is my crappy little app, and thousands of people started downloading it.

I didn't get any money from them, mind you, but at least I knew that thousands of people actually cared about voice cloning.

And then the other cool thing is there's a forum where these people download these apps and they leave feedback. So this guy here, Dominovic, he was a little bit upset at the developer who built the app, me, I didn't let him change the location of where the models got saved. So because everything runs on your own computer, all the models download on your computer, and it defaults to the C drive, and I guess this guy's C drive was full, and he wasn't very happy.

But I ended up adding the feature, and I think eventually he did leave a follow -up comment after he pirated the latest version saying that now you can move models to a different location.

Takeaways from Two Years of Building with AI

So, some takeaways, I guess, from my past two years of building software with AI. I think I fully believe that software is completely solved. So there isn't anything that you can't imagine that you can't build right now with AI.

And sure, I might have a leg up because I've been coding for most of my life, but my co -founder and my fiancé, who used to be a product manager, has actually taken over developing basically everything I worked on.

It's no longer I'm the developer and she does marketing or she does something else. She comes into every single code base that I have and she's productive and she's able to add new features.

And also today Anthropic released their Mythos model and I haven't been using it but I used Opus so if Opus was good enough, the new one is obviously even better.

When building features and when thinking of scope, like what to build, I think doing what What was typically considered impossible or just untenable is now the standard. The bar for quality is now exponentially higher than before.

And there's no such thing as, oh, wouldn't it be nice if we had a Mac OS app? Oh, it will take a year to develop. No.

Now you spend the one week that it will likely take to port over all your code, and now you have a Mac OS app, or you have an Android and iOS, or you have anything you can imagine. This is the new target. target.

You have to do the impossible.

And as a result of this, QA is way more important. If you're now shipping features that would have normally taken a year in a week, you need to QA the crap out of your apps.

And that now becomes the most important thing.

Shipping Faster Means QA Matters More

Moats Are Shrinking and Differentiation Gets Harder

And then the final thing is moats are drying up.

An investor once told me some number of years ago when I was trying to raise money that he basically asked me, like, if I can give a million dollars to some Stanford students in the US, can they replicate your app? Because if the answer is yes, then I'm not going to invest in you.

Now the answer isn't even do you need a million bucks, or the question isn't do you need a million bucks, you just need a large language model, and you need the software that somebody else has made, and you can reverse engineer their entire stack.

Algorithms, IP, all of this is drying up so quickly, and it becomes so easy to replicate replicate it, that differentiation becomes, well, I don't have the answer, but it becomes way more difficult.

And open source is a really cool way to see what people are working on and even if you don't understand the complexity, again, it doesn't matter.

There's somebody right now working on something probably tangentially related to one of your products and they have it on GitHub and now you can just go in there and ask Claude or or ChatGPT to explain it, and now you have that same thing inside your own product.

Conclusion

And that's it. Thank you. Any questions?

Q&A

Yes.

Preferred Tools: Claude, Claude Code, and the Shift Away from UI Builders

Claude. So, Claude and Claude Code. I've been through the gamut of applications.

I was a big fan of V0 when it first launched. This is like a nice UI, tell it, you know, build me some app and then it'll do it for you.

But now as models are getting smarter and smarter, all you need is access to the best chat GPT or Claude model, and then something like Claude code or Codex, sorry, yeah. Yeah, it's a lot easier.

Iteration Loop: How Often to Test and Adjust

How iterative is it when you're using Claude? How often are you, you know, checking if it works, going back, making the adjustments?

Extremely. And that's kind of a problem that I think a lot of developers were put off of, because you'll tell it to build something and it won't work, right? And so most people will be like, oh, well, if you can't do this, obviously you won't be able to do anything 10 times more complicated.

But that's not true. So, yeah, the first time you ask it to build you the macOS version of your app, it's not going to succeed. And then you just have to keep iterating. But the iteration loop is so quick that it doesn't matter.

It's just like another couple of hours, and then you have a working product.

Getting the First Users: Windows Store, Reddit, and SEO

How did you get your first few users? I mean, aside from, like, building a great product and putting it out there, Like was it difficult to find you know that get that initial traction, and you know how did you do that?

unfortunately I For voice creator I happened into a space that is very hot right now voice cloning. I haven't done anything I made a website I Uploaded the app to the Windows Store, and that's it.

I made a few reddit posts That's been my strategy for the the past apps that I've worked on With reddit you have to be a little bit less focused on marketing and more provide useful information Something related to the tech and then at the end by the way, you know, check out my app

But besides these two things I tried Google Ads Got rejected because my budget was too low so it's just been SEO blogs make some posts on reddit and

Yolo Wow, very impressive. I'm curious and

Revenue and Plans for a SaaS Version

How much revenue have you made net with this voice app specifically? So we launched in March. And we've been averaging about $3 ,000 a month since. And this is for the desktop version.

I'm also working on a SaaS version as well.

That's amazing. Thank you. Thanks, Robert, for the presentation. Really great initiatives.

Competing with Free Alternatives

I have a question on the voice. So there are other alternatives right now. Some of them are completely free. You already know which ones I'm going to talk about.

Even Canva provides some of the tooling, voice box, all that. So they work similarly, and you don't really need no credits, no nothing. You can start free. There might be limitations.

Maybe you might have more personified voice, better tuning or whatever, but why should someone choose Voice Creator Pro or the free solutions?

Dude, the honest answer is if you know about VoiceBox, you should probably use VoiceBox. So VoiceBox is a good one. Great piece of software.

software, what I did was I went to the GitHub page, and I saw all the issues people were filing for VoiceBox. People were saying, oh, it doesn't work with my graphics card, oh, wouldn't it be great if you could add this?

And then I just added all those things to my app. But if you just need basic voice cloning, yes, there's a thousand options.

So does the, I didn't, maybe I missed the slide, you didn't show the demo or something, Does it simplify the workflow for those who just want a couple of clicks and some description and they want to base out the voice depending on certain characteristics, they can quickly do that?

Yes. That's exactly it. Yeah? The vision is one click to set everything up so you don't have to do anything

yourself. Okay. Thanks. Thanks.

Hi. I'm here. Sorry. Thanks.

Privacy and Security Considerations

So I was just wondering, especially for your watch application, as I'm not that familiar with web coding, how did you check in terms of privacy and security your application has a it's really dangerous, right? So how did you check and what kind of guardrails are in your application regarding privacy Privacy as security.

So the watch app has two methods of communicating with AI. I tried to make it privacy first. So the core way that the watch communicates is through your own server.

So if you have a server at home that's running a language model and you expose it, you can connect it to the watch and you can talk to the LLM running in your basement to keep everything secure. here.

On the other hand, I am providing like a free chat GPT access. And so if you mean in terms of security, like where does your data go? If you use your own server, then it goes onto your own

server. If you use the free version where I'm providing free AI credits, then it goes off to some server that does the processing. Unless I misunderstood.

Cryptography or like are you using some key in terms of cryptography like as a security framework like do you have you don't have a security process like is there a login?

No it runs completely locally I mean I even made it so you can download a model onto your watch you don't even have to go to the cloud yeah I wouldn't recommend it but it works

Keeping Code Lean While “Vibe Coding”

So since you're live coding, how are you optimizing your code to minimize bloatware? Periodically. You tell Claude, hey, Claude, can you simplify my code?

Yeah. Yeah.

Hi.

Advice for Beginners Using AI in Everyday Workflows

So first of all, thank you so much for the demo and everything. thing and my question to you is not about your product it's about your experience so for me like I'm not sure about everyone but most of us or many of us would be some some like trying to

learn AI putting AI into the use of their daily workflows making things easier for them not making products so like in your opinion or in your expertise what would be the right path to maybe be like begin and someone who is a beginner and for someone who is an intermediate person how how should they proceed towards trying to resolve problems that are like in their

workflows rather than focusing on like like a lot of noise around the platforms because like a lot of people are doing so many things with ai but when you are at the beginning phase and you you see the final line of someone else, it feels like you are way behind, right? So what is your opinion on that aspect?

My answer is going to be very boring. If you would have asked me like a year and a half ago, I would have said, oh, yeah, use this tool, use this, this, this, harnesses, everything. Now none of that really matters.

Again, get a subscription to Claude or ChatGPT. key, tell them your problem, and it will find a way to fix it for you. There's no magic. These models have gotten so intelligent and so capable that you can literally just ask

them to do something, assuming it's digital, like it's on your computer, and then they'll just do it. And at worst, if they can't, they'll say, hey, I need you to download this software for me, or I need you to install this tool, and then it'll do it.