Case Study - AI-powered speaking coach transforming language learning

The Challenge

Traditional language learning apps follow a one-size-fits-all approach. They don't adapt to individual goals, learning contexts, or real-world needs. Users struggle with motivation, receive generic feedback, and fail to see connections between lessons and their actual communication challenges—whether that's professional presentations, casual conversations, or specific scenarios.

The vision for Saywell was different: an AI-powered speaking coach that personalizes every aspect of the learning journey based on how people actually want to communicate.

My Role: Creator, Architect & Entrepreneur

I built Saywell from the ground up—from initial concept through to App Store launch. This encompassed technical architecture, AI implementation, product design, and go-to-market strategy.

Technical Architecture

Designed and built a comprehensive TypeScript monorepo spanning:

• Web frontend (Nuxt 3/Vue 3) with modular design system
• Mobile app (Capacitor/Ionic) deployed to Google Play Store
• Backend API (Fastify/Node.js) with PostgreSQL and Prisma ORM
• AWS S3 integration for audio processing and storage

The architecture supports multiple learning modalities: Word of the Day, image descriptions, interactive cloze tests with drag-drop, conversational practice, and real-time audio recording—all built to scale.

AI Implementation

Created a sophisticated AI pipeline integrating multiple models:

Intelligent Journey Generation: Built an AI curriculum builder that analyzes users' learning goals (text or voice input) and automatically generates personalized multi-day learning journeys tailored to their context and objectives.

Voice-First Onboarding: Users can speak their goals instead of typing. Whisper API transcribes in real-time, GPT-4 extracts learning objectives and contexts, then generates a complete personalized curriculum—all while showing users their goals transform into tags with beautiful blur-to-focus animations.

Multi-Dimensional Evaluation: Developed a comprehensive AI evaluation system analyzing:

• Vocabulary richness and appropriateness
• Speaking pace (words per minute with consistency tracking)
• Pronunciation and clarity
• Tone and formality matching

Adaptive Content: Built systems to dynamically generate activity variants and conversation scenarios tailored to user proficiency and interests.

Contextual Feedback: Engineered prompts that provide encouraging, constructive feedback that feels human and supportive—not robotic or discouraging.

Product Strategy & UX

Behavioral Design: Implemented a "3-minute daily habit" model that reduces friction while maintaining engagement. Users practice in bite-sized sessions that fit into busy schedules.

Puffie: AI Character-Driven Experience: Created an animated AI companion that guides users through their learning journey. Puffie transforms what could be a sterile educational app into an emotionally engaging experience—offering encouragement, celebrating progress, and providing personality throughout the learning flow. This character-driven approach was an experiment in weaving narrative and personality into otherwise routine language practice, making users feel accompanied rather than alone in their learning journey.

Mobile-First Experience: Prioritized native mobile interactions—recognizing that language practice happens on-the-go—while maintaining web accessibility.

Micro-Interactions: Crafted delightful details like live recording duration counters, rotating motivational messages during AI processing, and smooth navigation transitions that mask loading states. Every interaction feels polished.

Entrepreneurial Execution

Market Positioning: Positioned Saywell as an "AI Speaking Coach" focused on confidence and real-world application—not just another vocabulary app.

Technical Foundation: Built extensible architecture allowing rapid iteration and new feature deployment without rewrites.

Data-Driven Development: Integrated Mixpanel analytics and Sentry error tracking from day one to understand user behavior and optimize conversion funnels.

Go-to-Market: Launched on Google Play Store with strategic content marketing emphasizing emotional benefits (confidence, clarity) over technical features.

Technical Highlights

AI Pipeline Architecture:

User Input (Text/Voice) 
  → Transcription (Whisper) 
  → Goal Extraction (GPT-4) 
  → Journey Generation 
  → Personalized Activities 
  → Real-time Feedback (Multi-model Evaluation)

Key Innovations:

• Transparent AI: Users see exactly how AI understands their needs through animated tag extraction
• Smart Journey Persistence: LocalStorage caching for instant navigation
• Progressive Enhancement: Graceful error handling for API failures
• Background Processing: Visible progress indicators that transform loading states into delightful experiences

Technology Stack:

• Frontend: Nuxt 3, Vue 3, TypeScript, TailwindCSS, Storybook
• Mobile: Capacitor, Ionic Framework
• Backend: Node.js, Fastify, Prisma, PostgreSQL
• AI/ML: OpenAI GPT-4, Whisper API, custom prompt engineering
• Infrastructure: AWS S3, Render.com, Clerk Auth
• DevTools: PNPM workspaces, TypeScript strict mode

Impact & Results

Product Achievements:

• Complete end-to-end language learning platform from concept to App Store
• Multi-activity system with extensible architecture for rapid feature addition
• Voice-first onboarding dramatically improving completion rates
• Sub-second API responses despite complex AI processing through strategic caching
• Zero-to-production in modern monorepo with 9 interconnected packages

Technical Achievements:

• Full-stack TypeScript monorepo supporting web, mobile, and API
• Comprehensive AI integration across transcription, evaluation, and content generation
• Scalable architecture with JWT authentication and stateless API design
• Robust error handling and analytics instrumentation

Key Learnings

On AI Integration: Prompt engineering for consistent, encouraging feedback requires extensive iteration. Voice input dramatically improves onboarding completion, but demands careful UX around recording states. Background AI processing with visible progress indicators builds trust.

On Product Development: Beautiful micro-interactions transform "loading states" into delightful experiences. Personalization at onboarding creates immediate emotional investment. Mobile-first constraints (3-minute sessions) drive focus and higher completion rates. Character-driven UX (Puffie) creates emotional connection and reduces the sterility often found in educational apps—users return not just for learning, but to interact with their AI companion.

On Entrepreneurship: Building full-stack enables rapid iteration without coordination overhead. Technical flexibility supporting multiple activity types allows pivoting based on user feedback. Early instrumentation is essential for informed decision-making.

This project demonstrates the full spectrum of modern product development: from AI architecture and full-stack engineering to product strategy and go-to-market execution—taking an idea from concept to market-ready product.