From Anki to My Own App: How AI and React Native Built VibeLing for Language Learning
A developer frustrated with the limits of existing language learning tools created VibeLing, his own mobile app. In just 5 months, the project drew over 1,000 users without any marketing spend. The key tech stack? React Native, Node.js, and AI models for generating examples and translations.
Problems with Existing Solutions
The author tested various language learning methods and pinpointed two main app types that fell short:
• Dictionary apps (like SkyEng) – they offer fixed word lists but don't let you add custom phrases or contextual expressions.
• Flashcard apps (such as Anki) – they require manual data entry, turning material creation into a tedious chore.
This pain hit hardest when studying Serbian, where user-friendly tools are scarce. That sparked the idea for a custom solution.
VibeLing Architecture and Tech Stack
For a quick MVP, the team chose this stack:
- Frontend: React Native + Expo for cross-platform iOS and Android development.
- Backend: Node.js as a proxy to AI APIs and AWS text-to-speech services.
- Infrastructure: Vercel for deployment, RevenueCat for subscription management.
- Website: Next.js with SSR for SEO optimization.
AI handled automatic translations and example sentences, eliminating manual user input.
Building the MVP and Early Challenges
The first version was a simple web page with one input field to validate the concept. Once proven, a mobile version was built in days. Development sped up with AI tools:
• Cursor and Claude for code generation.
• Ready-made UI kits from Figma Community Templates for design.
A surprise hurdle was Serbian's dual Cyrillic and Latin scripts. AI models inconsistently mixed alphabets, and AWS TTS only accepted Latin. The fix? An algorithmic converter instead of tweaking AI prompts.
Launch and User Feedback
The initial iOS release uncovered bugs and edge cases missed in testing. Negative reviews provided crucial insights for stabilization. Distribution happened via:
• The author's Telegram channel (1,500 subscribers).
• Posts in Habr's "Self-promotion" hub, pulling in about 700 users.
A major scare: RevenueCat's Google Play integration glitched, auto-refunding successful payments. This sparked bad reviews and a potential account ban risk. Fixed with manual config tweaks.
Current Status and Future Plans
As of this writing, the app boasts:
• Over 1,000 registered users.
• 700 active users in the last 28 days.
• 5 paid subscriptions.
Key roadmap items:
- Boost AI stability, especially for rare languages.
- Streamline the user flow from adding words to mastery.
- Explore image generation for associative recall, balancing token costs.
Key Takeaways
• AI jumpstarts development but doesn't replace deep coding – neural nets nailed the MVP, but thorny issues like payments needed hands-on fixes.
• Distribution beats perfect code – a rough MVP solving a real pain point can gain traction through targeted posts.
• Solve your own problems first – if it's useful to you, it'll likely help others too.
• Multilingual support demands nuance – quirks like Serbian's dual alphabets call for clever tech workarounds.
— Editorial Team
No comments yet.