Me & Tazavec: 5 Things I Learned From My Failed AI App
Over a year ago, I shipped my first solo app on Google Play. It was called Tazavec. I managed to earn €0.20. Here's my reflection on that journey and lessons that I am taking away.
7 min readBranislav Mateas
Copy post as Markdown
On this page
Some time ago, I shipped my first solo app on Google Play. (not really counting Cubio), I earned €0.20. Here's my reflection on that journey.
Note: Mobile version looked much better. This was a way to support non-Android users.
Picture this: it's early 2024, and AI is the thing. Looking back, it was nothing compared to today - but back then, it felt massive. I wanted to get my hands dirty with some API calls.
I needed a use case to practice on. Something simple enough to build, but real enough to actually ship.
The idea came from a simple observation: if you're the type who struggles to break the ice with strangers, or you want to go deeper on a date, or you just want to play truth or dare without the cringe - having the right question ready would help.
And I figured: if I'm doing this anyway, I might as well learn how to publish an app on Google Play myself.
So I built Tazavec. I would love to say vibe-coded but the code output back then was diametrically different and barely useful for the project.
Back to the app itself. You'd choose a depth from 1-5, optionally input a topic, and the app would generate conversation starters using AI. What started as "let me learn how LLM APIs work" might spiral into a real side project - or so I thought.
I even set up AdMob (a video ad every 5 questions, nothing aggressive - I just wanted to cover the AI token costs), published it on Google Play, and waited.
Total earnings: €0.20.
Late 2025, I killed the app. Looking back, here are five things that stuck with me.
1. I was lacking ambition
Here's something embarrassing. When I needed 20 testers to move from internal testing to beta release on Google Play, I messaged people like this:
"Hey, I'm just testing this thing, would you mind trying it?"
No pitch. No excitement. No "this could actually help you."
I'd built something I genuinely believed could work - the competitors had massive audiences (200k+ on IG), and honestly? Their apps weren't that special, mostly for iOS. Card games with generic prompts. Nothing personalized.
But when I wrote to people, I sounded like I was apologizing for existing.
Then a few people from the Yes Theory community said something that stuck with me:
"Try to aim higher. Write to that big YouTube channel - maybe they'd even collaborate with you. The worst they can say is no."
They were right. I eventually crafted a real pitch - the problem, the solution, why it's better than competitors.
But I never sent it.
Deep down, I didn't believe Tazavec was production ready. And it honestly probably wasn't. Which brings me to the next thing.
2. I built a house on someone else's foundation
As mentioned, when I started Tazavec, I just wanted to test API calls to LLMs. I picked Groq (not to be mistaken with X's Grok) as an ultimate solution - a low-cost inference provider - and used a Flutter package that abstracted their API. Connected it, shipped it.
Then one day, the model just... stopped working. No warning. No error message that made sense. Just broken.
I had no monitoring. No fallback. No way to know it was down until someone complained.
And there was a bigger problem I hadn't thought about: what happens when you scale?
Groq offered free API calls for small usage. But if Tazavec had actually blown up, I'd have been scrambling to figure out licensing, terms of service, pricing - all while users were already expecting it to work.
Looking back, the architecture should have been different from day one: pre-generate 1,000+ questions offline, use AI as a supplement not the backbone, build redundancy before you need it.
If your app's core feature depends on a third-party API you don't control, you don't have a product. You have a demo, or as we call it in 2026 - a wrapper.
3. Not everyone wants "smart"
I promoted Tazavec in self-improvement and travel communities - Yes Theory fan groups, that kind of vibe. I expected tech enthusiasts.
What I got instead surprised me.
A huge chunk of people were actively asking for an app that doesn't use AI. They wanted something simple. Offline. No data sent anywhere. Just good questions. Further proving my "wrong-architecture" narrative.
At first, I was confused. Why wouldn't you want the latest tech?
"Powered by AI" isn't always a feature. Sometimes it's a liability.
I could go on wrong architectural decisions on another tangent, but some include - how do you prevent serving user different questions to his topics? Especially if one changes AI model?
4. I understimated processes
Tazavec wasn't just an app. I was also running an Instagram account to promote it, which played a huge part in my opinion in the marketing "strategy" if you will.
I had templates. I did A/B tests on post designs. I genuinely believed in the product.
But here's the math that killed me:
40-60 minutes per post - if I wanted to be original. Meanwhile, competitors were pumping out 5+ posts a day with similar (or worse) designs. They just had systems. I had... me, doing everything manually.
Back then, n8n wasn't mainstream. Automation felt like overkill for a side project.
It wasn't.
I once heard: "If you compare yourself to the competition, you'll be at best as good as them."
Fair point. But in this case, being "as good as them" would have meant consistent output. And consistent output requires systems, not just effort.
5. I overextended myself
“The hardest part isn't building. It's maintaining.”
After about a year, I lost the spark. Between university studies and part-time work, Tazavec became something I was maintaining rather than actively developing. I'd fix a bug here, respond to a review there, but the energy to push it forward? Gone.
And I regret not setting up analytics earlier. I added Firebase a year into the project. A year! By then, I had no idea what users were actually doing in the app.
It didn't help that there's no easy way (as far as I know) to set up analytics on mobile like there is on web. On the web, I'd use Google Tag Manager as middleware - easy event tracking, no code changes needed. On mobile? Every single analytics ping had to be sent manually. Hardcoded. One by one. Which for an analyst, it's pretty uncomfortable if you have to ping the programmer with anything. Even if in this case, I was both.
I slightly regret not delving into mobile analytics a bit deeper. I know about Michal Nováček who's probably the most influential person in the space from my perspective - and even attended a webinar! I found out too late that it was focused mainly on iOS tracking, which for me building Android-first in Flutter is not that useful as you may have figured out by now. Still, it was a great insight and you should drop him a follow.
So... Is Tazavec dead?
Kind of.
The app is down. The Instagram is quiet. I earned my 20 cents and moved on.
But the idea? I still think there's something there.
With today's AI agents, I could probably rebuild and scale it properly. Different architecture. Better automation. Maybe even pivot it into something bigger - an app for long-distance relationships? A mood tracker with conversation prompts?
If you're reading this and you think you could do something with the idea - let's talk. Links are in the footer.
For now, Tazavec lives on my shelf of unfinished side-projects. But hey, at least I can publish on Google Play - the €20 license is forever. Take notes, Apple.
