About 2 months ago, I ended my 172-day streak on Duolingo. To be clear, this isn’t a “hate post.” Duolingo works for millions of learners. But my streak became a trap: I wasn’t learning Spanish, I was learning how to keep a streak alive.
And like most developers here, I ended up building my way out.
Context
I’ve always wanted to speak Spanish… but more than speaking, I wanted to understand it. I listen to reggaetón all the time, so I wasn’t starting from zero. But when a Bad Bunny song comes on, I wanted to feel the lyrics, not just hum along.
Enter Duolingo — the gateway drug for Spanish learners. I got hooked.
The Honeymoon & The Trap
In the beginning, it was fun. The gamification worked. My streak climbed. My vocabulary grew. But then the fear of losing the streak took over. Learning turned into an obligation.
Fast forward: I’m at a Bad Bunny concert in San Juan. The vibe was next level — but I only understood maybe half of what he was saying. I know the Puerto Rican accent is difficult but still six months of Duolingo everyday, and I still couldn't catch half of what he was saying (those tickets were not cheap)
That’s when I realized: I wasn’t practicing speaking. I was just answering quizzes.
Breaking the Cycle
The next day, I tried something different: I opened the ChatGPT app on my phone and forced myself into casual “Spanglish” conversations. Not full Spanish — that was overwhelming. But a mix.
I made some tweaks to my prompts:
- Cap responses at 2 sentences (so I wouldn’t drown).
- Mix English + Spanish depending on my comfort level.
- No judgment, no pressure.
It actually worked. For the first time, I was talking in Spanish consistently every day for a week.
But soon the builder in me kicked in. I wanted:
- A slider to control how much Spanish vs. English (my “Spanglish Level”).
- Adjustable voice speed (because listening is harder than reading).
- Less time prompt-tweaking, more time practicing.
And because I couldn’t find an app that did all that, I built one.
Building Lulu AI
I hacked together a prototype:
Frontend: Swift/ChatGPT Generated UI
Voice: Eleven Labs
Backend: Firbase DB/functions/Auth/Storage
The MVP was ugly, but it solved my pain point: I could finally practice without feeling in a small playground of my own creation.
Now it’s live in the App Store as Lulu AI
(not pitching — just sharing the journey).
What I Learned
Two months into building and using my own tool:
⚡ Cognitive load matters. Two-sentence replies keep me from mentally translating forever.
🎚️ Spanglish Level works. I started at 30% Spanish, now I’m at 80%. I’ll hit 100% eventually.
🎧 Listening is harder than reading. Even at slow speeds, audio comprehension lags behind reading comprehension.
⏱️ More speaking minutes. I’m practicing way more than I ever did with Duolingo — and I feel the difference when I catch lines in songs I used to miss.
Final Thought
Most developers (myself included) fall into the trap of building our own tools instead of paying for or adapting to existing ones. Nine times out of ten, that ends up being a distraction — and often leads to unfinished side projects that don’t solve the problem any better.
But sometimes the issue isn’t that the tool is bad — it’s that it wasn’t designed for you. For me, Duolingo’s streak mechanic killed my motivation. Building an alternative was the only way to align the learning process with how my brain actually works.
So now I’ll throw it back to you:
👉 When you’re frustrated with a tool, do you switch… or do you build?
Top comments (0)