Sol — Spanish Tutor

Project Overview

Sol is an AI-powered Spanish tutor that lets you practice the way people actually speak — by talking. Hold the mic button, speak in Spanish or English, and Sol transcribes your speech, replies in character, corrects mistakes with humour, and reads responses aloud with a natural Neural2 voice.

The app runs on phone, tablet, and desktop browser from a single React Native / Expo codebase, with platform-specific audio handling on native (expo-av) and web (MediaRecorder).

Key Features

• Push-to-talk — hold the mic to record, release to send
• Bilingual input — Whisper auto-detects Spanish vs English
• 70+ role-play scenarios — restaurant, airport, job interview, and more
• Smart corrections — gentle fixes for small mistakes, dramatic ones for big ones
• Neural TTS — Google Cloud Neural2 voices for natural Spanish playback
• Responsive UI — animated chat bubbles, typing indicator, adaptive layout

Architecture

User speech is recorded and sent to Groq Whisper for transcription and language detection. The transcript goes to LLaMA 3.1 with a structured JSON system prompt (Spanish reply, English translation, optional correction). The UI renders chat bubbles while Google Neural2 TTS speaks the Spanish response.

Core logic lives in src/services/ (Groq + Google TTS), with reusable hooks for recording and responsive layout. Scenario definitions are data-driven in src/data/scenarios.ts.