I couldn't get my daughter to clean her room, so I built an app where AI grades the before/after
Hey IH. Solo dev, first project I've actually shipped, so I'm a little nervous posting this.
It started with my daughter's room. I tried asking nicely, a chore chart, small rewards, negotiating, losing my temper. Nothing survived more than a few days. The one thing that clearly does motivate her is what her games do: XP. A number that goes up, and something outside her confirming she earned it. So I built her a referee for cleaning.
You photograph the mess, clean, photograph the result. Claude Vision compares the two shots, scores the improvement 1 to 10, awards XP and tells you what's still wrong. Around that: levels, streaks, 19 achievements, a leaderboard, and a feed where friends react to each other's before/after.
Then it started working on me, which I hadn't planned for. A todo item that says "clean the kitchen" sits there forever. A 6/10 with a note that the counter is still cluttered actually gets me up to finish it.
Some things I got wrong along the way.
The hard part wasn't the AI, it was making the app feel calm. Pages flickered on every navigation and it took me embarrassingly long to see why: the platform's own slide animation and my own fade were fighting each other, so a page arrived fully visible, dropped to zero opacity for a frame, then faded back in. That whole class of bug is invisible in code review and obvious the second you hold the phone. Stack, for anyone curious: .NET MAUI for one Android and iOS codebase, a C# backend on Railway, Supabase, Claude Vision for the judging.
Streaks I got wrong twice. A daily cleaning streak dies around day three and a dead one demotivates worse than none, so the main streak is moving to weekly. The code side is still open: production showed a streak of 1 for a user with no cleanings behind it, a local time versus UTC mess I patched without being sure I found the cause.
The thing I'm worst at is distribution, and I went in with no audience and no experience. My first plan was Polish Facebook groups about cleaning and home organisation, since people there are already trying to make themselves tidy up. Then I joined and read what actually gets posted: mostly cleaning companies advertising and people asking how to get a stain out, not communities that want a motivation app. My best channel went to almost zero before I posted a word. I expect to be wrong about the next channel too, just faster.
The app is going into closed beta now: Google Play on Android, TestFlight on iOS. One caveat on timing, because Google's rule shapes it. A closed test needs the group to start together and stay opted in for 14 straight days, so Android invites go out in one wave in early August rather than the minute you sign up. iOS is immediate.
The first 50 testers who stay active through the 14 days get Premium for life and a Founding Tester badge in the app. It's free, I'm not selling anything, I just need real people with real mess.
https://ordoapp.net/beta?source=indiehackers
And if you've found your first users with no audience to start from, or run an app through Play's closed testing, I'd take any advice.
I build an alarm app (WakeWin, mine) where one way to switch the alarm off is photographing an object you registered in advance, and a model confirms it's really that object. Same mechanic as yours doing a different job, so a couple of things you'll run into sooner than you'd like.
The scoring problem is variance, not accuracy. Two rooms that look about equally tidy have to score about the same or the number stops meaning anything. A 6 one day and a 9 the next for the same effort reads as the referee being unfair, and that lands harder on a kid than an adult, because a parent saying "not finished yet" can be argued with and a score can't. I'd spend the effort making the judge boring and repeatable before making it clever.
People also learn the frame. Mine get photographed through another photo, or the object quietly migrates to the bedside table. Yours will get the before shot taken after a bit of extra mess has been added, and the after shot cropped to the one clean corner. Better prompting doesn't fix that; constraining what the camera has to include does - same shot both times, with something fixed in frame that can't be moved.
benzbar already flagged the adults-are-a-different-user thing so I won't repeat it, but one consequence worth thinking about: the social feed is the first thing that breaks across those two. A kid will happily post a before/after to her friends. An adult will not photograph his actual kitchen for a leaderboard, and that's the feature you'd have built for the wrong one of them.
On channels, for whatever a small sample is worth: Indie Hackers is the only one that's produced real conversations for me. Reddit gives me views and nothing else, and Quora returned 51 views across 20 answers before I stopped. Your Facebook groups finding is the same lesson arriving earlier, which is the good version of it.
Thanks for such a detailed reply. Your app has a really interesting concept.
I’ve also run into the issue of inconsistent AI scoring, and I agree that it’s one of the biggest challenges. To deal with people gaming the system, I added a reporting feature. I also added privacy controls, so users don’t have to share photos of their mess if they’d rather keep them private. At the same time, I think there are still plenty of people who wouldn’t mind sharing those photos.
How did you get past Google’s testing requirements for your app?
Can't help on Google, sorry — WakeWin is iOS-only, Android's just a waitlist, so I've never run a Play closed test. Apple's the only review I've been through, and 2.5.1 got me: a feature reading Health data that wasn't core, and I couldn't justify to a reviewer why an alarm clock needs sleep records. Worth checking your photo permissions against that bar before you submit.
On the privacy controls — if photos can be private, the feed loses the people who'd benefit most from being watched, the ones embarrassed by the mess. You may end up with a leaderboard of people who were already tidy.
For first users, I would not pick a channel yet. I would turn the 14-day closed test into the channel test: tag every tester by source, then measure day-3 return, completed cleanings, and whether they invite a friend. The niche that produces 10 active testers beats the niche that produces 100 polite signups.
Yes, it’s a good idea, but it comes with its own challenges. Since many platforms only allow direct links to Google Groups and prohibit links to other external resources, gathering accurate information can be quite difficult.
The Facebook group part stood out because you killed the channel before even posting.
What are you using now to decide where the next 50 testers are most likely to come from — people already trying to clean more, parents dealing with chores, or people already motivated by gamified habits?
Hi Valentin, Your Ordo test already has the right shape: a defined cohort, a 14-day period, and a clear requirement for active testers rather than vanity signups. I’m building Build Before 2030: a manually reviewed public record where makers claim a Founding 100 number, set one measurable 30-day milestone, and later add proof of what shipped. The first cohort is free. Ordo feels like a strong fit. Interested? https://buildbefore2030.com
"Ha — this is hilarious and genius. How did you validate that parents actually wanted this before building it? I'm building Rallynex and trying to figure out how to validate early."
The detail I keep coming back to is that it started working on you, not just your daughter. A todo that says 'clean the kitchen' sits forever, but a 6/10 with specific feedback on the counter gets you off the couch. That's a different user entirely - adults who already want to do the thing but need an external referee to make the effort feel real. For what it's worth, the same dynamic shows up in fitness apps: the people who stick aren't always the ones who needed to be convinced, they're the ones who needed the number to confirm they actually did something.
Honestly? No idea. I have doubts about all three.
The Facebook thing spooked me a bit. So instead of picking a theory now and defending it, I'd rather try them one at a time and see what happens.
And right now I'm mostly just hoping the people who joined today are still there in
two weeks.