Overview

Built the full-stack mobile application for SpeakPing — an AI-powered language learning app focused exclusively on speaking practice. Unlike traditional language apps that emphasize reading and grammar drills, SpeakPing's core loop is designed around a single daily habit: answer one AI-generated question out loud in 30 seconds, then receive instant AI feedback on fluency, grammar, vocabulary, and pronunciation. The app supports 7 languages (English, Vietnamese, Spanish, German, Japanese, Korean, Chinese) and uses on-device Whisper AI for voice transcription — ensuring complete privacy with zero cloud upload of voice data. Available on the App Store with a freemium model.

My Role

Full-stack Developer — Designed and built the entire product end-to-end: React Native mobile app, NestJS backend API, Next.js landing page, and React admin dashboard. Handled architecture, UI/UX, AI integration, App Store deployment, and go-to-market strategy.

Key Features

Daily Speaking Loop: A frictionless 30-second daily practice — one AI-generated question, one voice recording, instant multi-dimensional feedback. Designed to minimize friction and maximize habit formation.

• AI Feedback Engine: Instant scoring across four dimensions — Fluency, Vocabulary, Grammar, and Pronunciation — with specific improvement suggestions generated per response.
• Progressive Hint System: Three-level scaffolding for users who get stuck — Outline → Key Words → Full Answer — reducing speaking anxiety while encouraging independent production.
• Streak & Gamification: Daily streak tracking with heatmap visualization, streak freezes, and gentle accountability mechanics to sustain long-term engagement.
• Multi-language Support: 7 languages available (English, Vietnamese, Spanish, German, Japanese, Korean, Chinese) with language-specific AI question generation and feedback calibration.
• Freemium Monetization: Free tier with 4 daily questions and ads, Pro tier ($1.99/month or $14.99/year) with unlimited questions, ad-free experience, and streak freezes.

Tech Stack

React Native, Expo, TypeScript, NativeWind, NestJS, PostgreSQL, TypeORM, Whisper AI (on-device), Next.js 15, Tailwind CSS, React (Admin), Vite, TanStack Router, Radix UI

Challenge & Solution

The core technical challenge was implementing accurate speech-to-text transcription across 7 languages entirely on-device. Cloud-based STT solutions (Google Cloud Speech, AWS Transcribe) offered high accuracy but required uploading user voice data — conflicting with the privacy-first architecture. I integrated the Whisper AI model to run directly on-device, eliminating network latency and cloud dependency. The key tradeoff was managing model size vs. accuracy across languages — tonal languages like Vietnamese and Chinese required careful tuning of the transcription pipeline to handle pronunciation nuances that directly impact the AI feedback accuracy.

Links

Website: speakping.com

App Store: https://apps.apple.com/app/speakping/id6757438993

Screenshots: