GTM Sauce

Pushcroll · B2C · push-ups to unlock social media

Rebuild push-up tracking to work from the front so faces show on video

Unknownproduct-viralityorganic-short-video

What they did

Push-up detection (Google MediaPipe, on-device for low latency) was first built side-on, as in most demos. They rewrote it to detect from the front and customized the skeleton overlay so the face stays visible, because front-facing reps with a visible skeleton are more appealing in creator and organic videos. Alejandro argues every consumer app needs a visual element for these platforms (Cal AI's calorie tracking, Quitter's streak).
Stage
MVP to v2

In their words

Mario Manero & Alejandro Sanchez - Our App Makes $30K/Month Profit Using This Simple Strategy (copy us)

Play from 11:23

Mario Ortiz Manero… that I was building. So, I literally just like mostly imported most of it into this new sort of format, and then slapped AI detection into it. We had to change a lot about it, of course, and then like the issue is that I I started building the other app with Android instead of iOS. So, we first had to do like ship Android, and then iOS. But yeah, like we met up, and then we we decided how to approach it and and like spend some days building it.

10:16

Alejandro SanchezYeah, so pretty much the first version is very simple. You just select the app you want to block, like, let's say, X or Instagram, and you have a button, "Do Push-ups." And you you see your budget. And I think like actually the first version was even a sideways detection. And then, we did switch to a front detection, because it's more visually appealing. And yeah, that's pretty much it, like just this these two screens pretty much. That's the app.

10:44

Mario Ortiz ManeroSuper scrappy, like based on on demos that we found online with like AI detection and so on. And then, we had like the the core feature is like the AI detection, right? And so you really have to nail down how well that works. Um so we had to do like lots of like put a lot of effort into making it better, but the first version just worked, so that was good enough.

11:02

Joseph ChoiThat's crazy, this whole app is just two screens, I guess plus one uh you know intro screen. But then there's a few things that go into those two screens. Like the first one, there's the tracking where you it tracks how many minutes you've scrolled on each of the apps. And then the second screen, obviously there's the AI. Was it hard to build the AI? Like what went into that?

Mario Ortiz Manero… you could do push-ups. Uh that that was it. And then, like when you opened one of the blocked apps, you got a block screen based on like how much budget you had. So, like AI detection, getting it to work just barely was fairly easy, cuz there's like push-up detection, in particular, is one of the most common like examples of how to do visual recognition and and so on. So, it was pretty easy to get like a first version, but then it got harder as we tried to get it like to work for our use case, because in most cases, um they did it from the side, and we really wanted to do it from the front, for example, cuz it goes back again to that what we were talking about, like making the app viral. If the person is doing like the push-ups from the front, it really makes it easier to do influencer partnerships, cuz you just you see their face, it's a bit more appealing and so on. Sure, we had an initial version, but it was completely rewritten at some point, because we moved it to be from the front. Also, like we only did Android, right? So, at some point, we also wanted to do the iOS app, and had to like completely also change the framework and so on to make it work on both, sort of like in without having to code it twice.

12:38

Joseph ChoiInteresting, so is this just OpenAI, like OpenAI video?

Get tactics like this every Monday

New tactics from the week's founder interviews, each linked to where it was said.

More from this episode

Tags: visual-hook, demo-friendly, ai-skeleton