Subscribe
Share
Share
Embed
Interviewing indie founders about their journey and their products. itslaunchday.com
Dagobert (00:01)
Hey Emanuele, welcome to lunch day.
Manuele (00:04)
Hey, hi, how are you?
Dagobert (00:07)
Yeah, I'm good man. yeah, so I'm trying to make these about the stories of the founders and I've never met you yet. So I'd be very curious to see, know, well, like, Manuel, is that like Spain? Where are you from?
Manuele (00:24)
No, it's
Italian. It's Italian. I'm from Italy and yeah.
Dagobert (00:26)
Italy, awesome.
Lots of Italian indie makers. Lots of French ones too, by the way, actually. Yeah, that's interesting. So where are you based? Yeah.
Manuele (00:29)
Yeah, true, true, that's Yeah, yeah, Cousins, you know? Right now
I'm in Italy. I'll spend ⁓ for sure July and August here, and then we'll see.
Dagobert (00:43)
Yeah.
Okay, you traveled a bit?
Manuele (00:46)
Yeah, I travel a bit, try to move around. But let's say that the bass is now here, so at least for a bit.
Dagobert (00:50)
Nice.
Yeah, awesome. And so how did you get, and right now this project you're going to show, like is it like, are you like full time in the hacking? Do you have a job on the side? How is that looking?
Manuele (01:05)
No, I'm a product designer and I run a digital design ⁓ studio ⁓ that's called Hauer Fires, but I'm the type of guy that always try to build ⁓ tools and software to support my own workflow. ⁓ started with, I mean, a while ago, the first thing that I built was a Photoshop plugin ⁓ that actually got, ⁓ it was something to create
Dagobert (01:12)
I
Yeah.
Manuele (01:34)
⁓ like browsable prototypes before they were a thing. The name was lasagna. Yeah, basically you can create PNGs in Photoshop and then connect like hot area that you can click and move in the PNGs. Yeah, yeah, yeah.
Dagobert (01:41)
Which kind of prototypes?
Yeah.
⁓ this interactive prototype like InVision did like 10 years ago.
Manuele (01:58)
Yeah, and I did the,
Dagobert (01:58)
Yeah, I see.
Manuele (01:59)
yeah, the plugin was even before that. So very rough, but kind of working.
Dagobert (02:06)
So you're a designer and
engineer at the same time. You're developer and designer at the same time, kind of.
Manuele (02:10)
Yeah, mean,
engineer maybe is a typical word, but yeah, I always, yeah, get my hands dirty with code, let's say, and try to build my own tools.
Dagobert (02:14)
Yeah.
And these
plugins, that's interesting, because I feel like it can be something where distribution is a bit easier maybe, because I feel like some people sleep on that. You can just build a plugin, and if it's successful, then I guess you have more visibility somehow. How does that work to distribute it?
Manuele (02:42)
Yeah,
mean, nowadays with ⁓ tools like Figma and Framer, they have this ⁓ community slash ⁓ app store, maybe something like an app store for the plugins and stuff. And maybe it's easier to get some kind of traction for your product.
Dagobert (02:56)
Yeah.
Manuele (03:02)
Back then it was not. ⁓ The first thing that actually got some traction was a plugin for SCATS to have images from Unsplash directly in the ⁓ software. Yeah, yeah, SCATS out. Yeah, yeah, So, yeah, a few years back...
Dagobert (03:18)
In Sketch, you said Sketch? Yeah, I use it, it's my favorite. I still use it, yeah.
But you made this plugin or like it's the, it's just an example you're giving, this Unsplash plugin.
Manuele (03:32)
No, no, I
made this plugin. That was the first one that actually got some traction on the market. launched it even, I mean, it was like they say hunted on product hunt. ⁓ Yeah. And the, yeah, got some, ⁓ yeah, traction for the product. ⁓ The name was Unsplash It. So maybe you use it, I don't know.
Dagobert (03:44)
Yeah.
No, I don't use this, but I think they have a default plugin now that does this. Maybe they took the ID from you, but it seems like there's two plugins by default, and one of them is Unsplash, I think.
Manuele (04:03)
Nowadays, yes. Yeah.
Yeah, because
it was really useful. But I mean, it was like a very simple idea because I know that few people were working on the same idea at the same time.
Dagobert (04:13)
Wow, okay.
Manuele (04:22)
⁓ For some reason I think that my plugin was the one that got hunted on Product Hunt and was used before before Sketch had their internal official collaboration with Unsplash. That by the way of course works way better than the plugin itself.
Dagobert (04:29)
Yeah, the first one.
Yes, he.
because they can have like complete control and everything for sure. Wow. Okay, so you're always was tinkering ⁓ and you have your design studio. Okay, so did you make many products? Like how many is it? How many parts did you build?
Manuele (04:44)
Yeah, of course.
Again,
most of them are for internal use. We always try, I say we because it's not just me, we have three, four people. We try to optimize processes, so we build our own tools, but every now and then, yeah. And again, we try to...
Dagobert (05:13)
And it's the same people in your design studio, the whole team? Okay, I see, see, awesome.
Manuele (05:23)
optimize our processes and so we build some tools to do that. ⁓ And sometimes the thing that I build I try to also publish them on the market on the site. But to be honest that's something that has not happened that often. ⁓ And yeah I mean the origin of the product that we are going to talk about today is actually a bit weird because I was working on something else for internal use.
Dagobert (05:35)
Yeah, for sure.
Manuele (05:52)
And then I had the need to have someone that actually was, ⁓ that been able to listen to what I say and transcribe it in text because I was working on a prompt manager and I was using Cursor to build the prompt manager. And so I realized that I was spending my days and typing, typing, typing stuff. And, and you know, when you...
Dagobert (06:02)
Yeah.
Okay.
Yeah.
Manuele (06:21)
try to do fast iterations. Even if you are very fast typer, it's, I mean, you speak three, four times faster. So, yeah, I mean, I think I'm, yeah.
Dagobert (06:31)
Yeah, it's true, I think I
use it most when my ideas are unclear, because AI is going to sort it out. when your ideas are... If you have a clear idea, maybe just typing one sentence, I still do that. But if it's trying to list things while you're still thinking about it, then audio is way better. Yeah.
Manuele (06:41)
Exactly.
Yeah,
true, true, true. Yeah, exactly. When you try to do fast iteration and something like cursor, having something that you can, I mean, press a keyboard, press a key on the keyboard and have the thing listening to you and then transform it into text. It's not bad. So, well, I created something fast for my own use. And then I thought that it was really useful. thought, I mean, there are some similar solution already in the market.
Dagobert (07:16)
Yeah.
Manuele (07:24)
But all of them, I think most of them, ⁓ actually take your audio, send it to the cloud, have some cloud process to transcribe and get text back to you basically. So you need to be connected. You of course need to pay a fee because you are using resources, cloud resources. And I thought that with I mean, with the machines that most of us have.
Dagobert (07:32)
Yeah.
Yeah.
Manuele (07:53)
nowadays, ⁓ like the series processor for ⁓ Macintosh. I mean, those are powerful machines and ⁓ Whisper, it's an open source model for transcription that is, especially with smaller model, very fast. And you don't have the need to send data online. You can do it on your machine.
Dagobert (07:57)
Yeah yeah, for Mac.
That's really awesome because I tried one of these ⁓ big ⁓ transcription things and that's the thing that's annoying is that, that's what you said, it's about being faster. But then because it has all these online processing, it's not much but this fucking one second delay before and after, that's not what I want, you know? And sometimes I type.
Manuele (08:39)
Yeah. Yeah, yeah,
Dagobert (08:43)
I didn't,
Manuele (08:43)
Yeah.
Dagobert (08:44)
so that's what's interesting about your product. And before this call, I was just testing that your screen sharing was working. And when you share your screen, it's like this big screen. And I'm like, it's too big. People are not gonna see anything because the resolution is too high. And you said, no, but it's super simple. They don't need to see anything. And I really liked that you said that. So can you show us now? it's a...
Manuele (09:09)
Sure, sure,
Be glad to do that.
Dagobert (09:11)
I
think it's, I really like that thing that you focused on speed, because that's basically the value. The value is to go faster. So it's so stupid to have all the network latency that you probably don't need, because transcribing, I mean, it might be a solved problem with a small enough, you know, local thing already. So it's solvable. You don't need the latest tech. you, it does, especially that the LLM,
Manuele (09:33)
Yeah.
Dagobert (09:39)
is okay with some little mistakes, he can understand. So as long as you get the idea, it's fine. Yeah.
Manuele (09:42)
Yeah, definitely. Yeah,
absolutely. Yeah. Yeah. So yeah, as I was showing earlier, basically, hard carry, it's always a keyboard shortcut away. So you just push on your keyboard shortcut that you picked. And as long as you have, you know,
So you have the ⁓ widget to appear on the bottom of the screen. And when you've done speaking, you just press enter and yeah, it's nice, right? And you have the thing that you just said in your, yeah, text field. It's super easy, but that... ⁓
Dagobert (10:18)
Yeah.
It's beautiful by the way.
Yeah.
Manuele (10:33)
Yeah, that was the idea from the beginning. I don't want to type, I just want to be able to push a keyboard, say something, press enter and have it out the output in the, yeah, here in the text field that it was.
Dagobert (10:48)
So you basically press a key,
it activates the thing, and when you're done you press enter.
Manuele (10:54)
Yeah, exactly. Yeah. The keyboard shortcut is you can pick the one that you prefer here, of course. ⁓ This also is important. From the application itself, you can decide which model to use. So you can test different models. can download and test different models and see what is the best for your style and your machine as well.
Dagobert (10:56)
I see, okay.
Yeah.
Okay.
all local models
that you can download.
Manuele (11:23)
Exactly. You download
it. You know, it's always a threshold, that, sorry, not a threshold, a trade-off, meaning that the speed of the application and its accuracy, it's depending on the model that you use. So I'm using a fairly large model, but still it's fast enough.
Dagobert (11:40)
Yeah, yeah.
Manuele (11:45)
And also the multi-language version of the models cover really a lot of different languages. So if you're not, if you're not intent to use it with English, you can. I do it with Italian all the time. So it's work well in Italian. By the way, by the way. Yeah. Okay. ⁓ Yeah. Let me close this. And yeah, that's basically nothing more.
Dagobert (11:53)
Yep.
Okay, and it's working. Okay, okay.
And
how do you position it? Because that's interesting. Because I think maybe positioning on speed and local first. Because I guess a big problem with all these tools is it sends your data to the cloud, privacy concerns. Did you use that in your positioning, in your marketing yet? Is that something you're considering?
Manuele (12:29)
Yeah,
I did. I think that the differentiator for this is the, I mean, how easy it is. ⁓ Of course, it doesn't have all the, like, buzz and whistles that all those ⁓ bigger solution have, but they, again, they send your audio to the cloud and does cloud processing and so on and so forth.
Dagobert (12:39)
Yeah.
Manuele (12:57)
And to answer your question, yes, I use the idea of having a privacy first solution. You pay for it one time. It's not something that you pay on a monthly basis because you don't ⁓ consume others resources. You just use your machine. ⁓ And then, yeah, you can use it how much you want and ⁓ whenever you want, even if you're in the woods or on a plane, it still works. So even if you're not connected, I mean.
Dagobert (13:05)
Yeah.
Yes.
And you know, that's how it should be. I think it's because all of this AI kind of revolution made us want like put everything in the cloud. And I think that's why also Apple kind of struggled when they said they were going to add AI features because they're very privacy focused and they're also wanting to be local first. So it made it way more complex because, but I think for some transcription, it's reach a stage that's good enough.
that you don't need to always update your model, always have the latest state of the art, because once it works, it works. It's not going to understand you better once it understands you. It's good enough. There don't need to be this race of always like, we have the latest model for transcription, so you need to pay monthly. It's bullshit.
Manuele (14:05)
Yeah, yeah, definitely, yeah.
Yeah,
and to be honest, even the smaller model are precise enough to be used in a day. every day. Yeah. I mean, I picked the largest one to be honest, because I wanted to do a first good impression in the demo. Yeah, it's a bit slower. Yeah. Anyway.
Dagobert (14:24)
Yeah, yeah, for sure.
Yeah, to be sure, just in case. But it's 99 % the same, I guess. Can you just
try, you know, we'll have a perfect demo on the website anyway. Can you just try like the slowest model? I mean, the quickest, just to see how fast it is. And I'm not trying to put you on the spot, but I'm just like, for me, I think speed is the most important. And I think it's okay if it makes a bit of mistakes, because AI will understand it anyway. So.
Manuele (14:45)
Sure, sure,
No, no,
Yeah, can
you see my screen or not? Okay.
Dagobert (14:58)
No, not yet.
Manuele (15:09)
Okay, so for example...
Yeah, I'm going to pick the smallest one. Yeah, this one in English only.
Dagobert (15:17)
Yeah, let's try this one. Fastest.
Manuele (15:27)
Okay, let's see how this works. ⁓ Does it understand what I say or not?
I mean, yeah, yeah, yeah, yeah, yeah, yeah.
Dagobert (15:35)
Nice, okay that was instant and that works. This is the thing, you know, because
when we spend all day in cursor, this fucking two second lag is the most annoying thing and you fixed it. That's it, man.
Manuele (15:43)
Mm-hmm.
You're right. Yeah, yeah, yeah. Yeah. I mean,
I tried this already. ⁓ To be honest, I tried the multi-language version. ⁓ Of course, it's not as precise as the largest one, but still usable and pretty much instant. So not.
Dagobert (16:03)
Yeah, and you know, even if it's just English, I don't mind speaking in English for coding. ⁓ Wow. This is cool, man. Yeah. I think I will probably... I want to bring every indie maker some sales. That's the goal of lunch day. Well, maybe this one I will pay myself because I really want to use it. Maybe that's how I will solve this problem. I will buy all these products myself.
Manuele (16:07)
Mm-hmm. Mm-hmm.
I'll
be glad if you give it to you so you can give me feedback. I still have something. I mean, there are some rough ideas here and there. so ⁓ as I did, yeah.
Dagobert (16:37)
Yeah, I get it. ⁓
yeah, just one question about this because I'm using an app like this. Can you easily choose the microphone you use or is it just default?
Manuele (16:48)
That is one of those rough edges that I was talking about. Yeah, yeah, yeah. I mean, it takes, you have a list of ⁓ input devices and it takes the first one, but you cannot switch, you cannot switch it from the application.
Dagobert (16:51)
Okay, that's what I was thought of. Okay, cool.
Okay, okay
I see, but you're planning to update it? You're planning to improve that, I guess. Yeah. Yeah.
Manuele (17:08)
Yes, yes, yes. There are some, ⁓
yeah. I mean, what I would like to do is to improve the output of it. ⁓ But to be able to do that, I need to take ⁓ what the description model gives back and send it to another lamp. And that, I mean, that's kind of not.
Dagobert (17:27)
Yeah.
Manuele (17:32)
in line anymore with the initial idea. And so I have several ⁓ idea for improvements. A lot of people told me that they want to have the application running in the background. ⁓ Right. Kind of, because right now, if you click outside, the application closes itself. Like it stops, stop listening because it's the idea was just to use it instead of the, of the keyboard, you know.
Dagobert (17:47)
like all the time and just whenever they speak.
Yeah.
Yeah,
yeah.
Manuele (18:02)
But ⁓ yeah, some users are actually asking if they can have a version of it that can stay there even if they click around and do other stuff in the, yeah.
Dagobert (18:13)
I've
done this with another app, I've done this because you know what happens sometimes is you have an idea and you're just like talking and while you're talking you think shit I need to check something online so you leave it on and you check something online. Also another thing is you know it automatically pastes where your cursor was you know that's what it does in the input selection you were but sometimes I start talking and I click around
Manuele (18:17)
Yeah.
Yeah.
Yeah. Yeah.
Dagobert (18:40)
And so I need to come back somehow, or I wasn't in the right input field. So before validating, I'm clicking on the right input to make sure it pasted at the right place. I'm just giving you some feedback because I use this thing.
Manuele (18:50)
Yeah, yeah,
yeah. You hit exactly the issue that I'm having right now in trying to developing this feature. Because when you stop the listening phase, the application actually, what it does behind the curtain, let's say, it takes the transcription. ⁓
Dagobert (19:00)
yeah?
Yeah.
Manuele (19:13)
place it in the clipboard and then use the clipboard to paste that in the text field that you have active. But if you click around, then that's not the active text field anymore. So exactly as I said. So I'm trying to find a solution for this issue right now.
Dagobert (19:17)
Yeah.
Yeah, that's it.
But that's awesome, I really think, yeah, the speed thing, at least for me that tried these apps already, it's really the speed that, like my feedback as a user, like yeah, speed is the number one thing and local is awesome because it's speed guaranteed, so yeah, that's awesome.
Manuele (19:52)
Yeah.
Dagobert (19:52)
Do
you have ⁓ other things you're working on? Or is it your full focus?
Manuele (19:57)
I'm working on the ⁓ initial product that actually sparked the need for this. It is ⁓ a prompt manager. I don't know how you actually right now, if you have some kind of issue like this, but I used to have this Google Docs with a lot of prompts copied there, the ones that I use the most, that I reuse basically. I don't know if you do something similar, but many people that I spoke with actually.
Dagobert (20:04)
Okay?
Yeah.
Yeah, I do this
for cursor. have a file with a couple of prompts I keep copying and pasting a hundred times.
Manuele (20:33)
Right,
Yeah. And so, yeah, good point. And Cursor was one of the use cases that I was focusing on, but I sold that one by having like a folder that I copy and paste in new projects for Cursor. mean, super simple solution, but it kind of worked. Yeah. Yeah, exactly. But I also have... ⁓
Dagobert (20:50)
Yeah.
Yeah, I do that too, yeah, I just have a prompt folder in cursor, yeah.
Manuele (21:02)
like prompts that I use and reuse in other LLMs ⁓ on charge PTE, on cloud, maybe to generate content or to generate PRDs for example. ⁓ PRDs like product requirement documents, like yeah, and documents that I use then in cursors as well. But I mean.
Dagobert (21:15)
To make what? PRDs?
yeah, okay, I see.
but you generate them from another LLM. Yeah, I do the same for the planning phase when I code a project. I do with Gemini, yeah, yeah, okay.
Manuele (21:32)
Yeah, yeah, yeah,
exactly. Yeah. So I have these prompts that I reuse all the time in a Google Doc that I use to store those prompts. So.
I thought it would have been nice to have a small software that can be a manager of those prompts. And when I need them, again, I can use a shortcut, have an interface similar to what Harker does, and select the prompt and use it. So yeah. ⁓
Dagobert (22:15)
just
a list of prompts and you can just have a shortcut to copy and paste and to just copy one immediately, just saving you one click. Instead of like click, select, copy, just click and it copies it and you can paste it. Like this tiny speed thing. You could make like a whole suite of like speedy prompting tools, like the voice that's instant, the copy paste thing. That can be, because that's the big value. Yeah. Yeah.
Manuele (22:26)
Yeah.
Yeah, exactly.
They work well together, yeah. Yeah, they work well together. And
something else that you usually do is to, at least many people does, elements in the prompt that actually you want to change. So yeah, right. So...
Dagobert (22:55)
Yeah, I have that too, yeah.
Manuele (22:58)
the thing that's usually placed inside square brackets, you You know, that part needs to change when I actually use the prompt. And so Gedia has to have that application to actually use select prompt. And then it says, okay, what do you want to place here and here and here? Enter, and you have the prompt compiled with the new information that you needed to add in the output. I mean, in the, whenever you want to use it.
Dagobert (23:03)
Yeah.
Yeah, I
totally see it. Yeah, it's all about removing clicks. Like you click once, then it asks you what you want to put in it. You type it or you speak it. And then you click another time and boom, you have your thing ready to paste. That's cool. Yeah, that's awesome, man.
Manuele (23:34)
Or you speak it.
Yeah. Yeah.
And so you
to answer your question. I'm working on Harker and this other application that if it comes out as I would like it to, I will also publish. Let's see how it goes.
Dagobert (23:53)
Yeah man, I wish you
good luck with that. So, well, know, really happy to have met you and to have discovered this product. I really like it. So, you know, I really hope Launch Day can bring you some sales.
Manuele (24:03)
Great.
Yeah, I hope it too. And thank you for having me here. It's great project as you're running.
Dagobert (24:12)
Thanks man. Have a good one. Cheers.
Manuele (24:15)
Yeah.
Cheers. Bye. Bye everyone.