Home page
Home Page 2
Home Page 3
Home Page 4
Home Page 5
Onboarding
Reading assistance
Writing assistance
Speaking assistance

LinguistLM - AI-Powered Language Learning Platform

Inspiration

Language learning has traditionally been expensive, time-consuming, and often intimidating. We've all experienced the anxiety of trying to speak a new language, worrying about making mistakes, or struggling to find practice opportunities that fit our schedules. We were inspired to create LinguistLM by the vision of making language learning accessible, natural, and confidence-building for everyone. With AI technology advancing rapidly, we saw an opportunity to create a personal language tutor that's available 24/7, never judges your mistakes, and adapts to your unique learning pace and style.

What it does

LinguistLM is a comprehensive AI-powered language learning platform that offers three core learning modes:

📖 Read Mode: Practice reading comprehension with curated texts, real-world articles, and instant translations. Build vocabulary and understanding at your own pace across multiple difficulty levels.

✍️ Write Mode: Get instant, detailed feedback on your writing - from casual messages to formal essays. Our AI analyzes grammar, vocabulary, style, and provides personalized corrections and suggestions.

💬 Speak Mode: The crown jewel of our platform - engage in real-time, voice-to-voice conversations with an AI language tutor. Using Google's Gemini Live API with native audio capabilities, users can have natural, flowing conversations where:

Your speech is transcribed in real-time
The AI responds with natural voice synthesis
You receive gentle corrections and encouragement
Conversations adapt to your proficiency level
Practice feels like talking with a supportive friend

The platform supports 50+ languages, from popular choices like Spanish, French, and Mandarin to less common languages, ensuring everyone can learn the language that matters to them.

How we built it

Frontend Stack:

Next.js 16 (App Router) for a modern, performant React framework
TypeScript for type safety and better developer experience
HeroUI/NextUI for beautiful, accessible UI components
Tailwind CSS for responsive, utility-first styling
Custom design system with carefully crafted color palette (llm-chinois, llm-sea-glass, etc.)

AI Integration:

Google Gemini 2.5 Flash for text-based reading and writing assistance
Gemini Live API with native audio for real-time voice conversations
Audio processing using Web Audio API for 16kHz input and 24kHz output
Real-time transcription for both user input and AI responses

Architecture:

Next.js API routes for secure backend operations
Client-side audio streaming with ScriptProcessorNode for microphone input
WebSocket-style connection via Gemini Live for low-latency audio exchange
Session management for maintaining conversation state
Environment-based API key security

Key Technical Features:

Real-time PCM audio encoding/decoding
Automatic audio buffer management for seamless playback
Turn-based conversation handling with interruption support
Responsive design that works on desktop and mobile
Dark mode support throughout the application

Challenges we ran into

1. Audio Streaming Complexity: Implementing real-time audio-to-audio chat was significantly more complex than expected. We had to:

Understand PCM audio encoding at 16-bit depth
Handle sample rate conversions (16kHz for input, 24kHz for output)
Manage audio buffer timing to prevent gaps or overlaps
Deal with Web Audio API deprecations (ScriptProcessorNode vs AudioWorklet)

2. TypeScript Compatibility: Getting the @google/genai package to work smoothly with Next.js and TypeScript required careful type definitions and understanding of when code runs on the server vs. client.

3. Real-time State Management: Coordinating between audio playback, transcription updates, and UI state while maintaining smooth user experience required careful callback architecture and React hooks optimization.

4. Responsive Layout: Creating a balanced, professional layout that works across different screen sizes while maintaining the visual hierarchy between conversation display and controls took multiple iterations.

5. Error Handling: Implementing graceful error handling for microphone permissions, API failures, and network issues while keeping the user experience smooth and informative.

Accomplishments that we're proud of

✨ Seamless Voice Conversations: We successfully implemented real-time, natural voice-to-voice conversations with an AI - something that feels almost magical to use

🎨 Beautiful, Cohesive Design: Created a polished UI with a custom design system that's both functional and aesthetically pleasing

🚀 Full-Stack Integration: Built a complete end-to-end application from audio processing to UI, all working together seamlessly

📱 Responsive Experience: The platform works beautifully across devices, from desktop to mobile

🔒 Secure Architecture: Implemented proper API key management and security practices

💡 User-Centric Features: Added thoughtful touches like auto-scrolling chat, real-time transcription, session statistics, and gentle error messages

🌍 Multi-Language Support: Built a platform that can genuinely help people learn any of 50+ languages

What we learned

Technical Skills:

Deep dive into audio processing and the Web Audio API
Real-time streaming architectures and WebSocket-style connections
Advanced React patterns with hooks and callback optimization
Next.js 14 App Router best practices
TypeScript advanced types and generic constraints

AI Integration:

Working with cutting-edge AI models (Gemini 2.5 Flash, Gemini Live)
Understanding modalities (text, audio) and how to leverage them
Crafting effective system prompts for educational contexts
Managing AI conversation state and context

Product Design:

Importance of user experience in educational tools
Balancing feature richness with simplicity
Creating an encouraging, non-judgmental learning environment
Responsive design principles and flex layouts

Soft Skills:

Time management in a hackathon environment
Prioritizing features (MVP vs. nice-to-have)
Debugging complex, multi-layer systems
Reading documentation and adapting reference implementations

What's next for LinguistLM

Short-term Goals:

Language Selection: Add ability for users to choose which language they want to practice
Difficulty Levels: Implement beginner, intermediate, and advanced modes
Topic Selection: Let users choose conversation topics (travel, business, casual, etc.)
Progress Tracking: Add detailed analytics showing improvement over time
Save Conversations: Allow users to review past conversations and corrections

Medium-term Goals:

User Accounts: Implement authentication and personalized learning paths
Vocabulary Lists: Generate custom vocabulary lists from conversations
Pronunciation Scoring: Add detailed pronunciation feedback and scoring
Multiple AI Voices: Let users choose from different voice options
Mobile App: Create native iOS and Android apps for on-the-go learning

Long-term Vision:

Community Features: Connect learners with similar goals
Gamification: Add achievements, streaks, and challenges
Cultural Context: Integrate cultural notes and real-world scenarios
Professional Certifications: Partner with language certification programs
AI Tutor Specialization: Different AI personalities for different learning styles
Offline Mode: Download lessons and practice without internet
Integration with Other Platforms: Connect with language learning ecosystems