Duolingo-V2

Duolingo meets AI voice chats. Practice speaking a new language by actually talking to an AI—and getting real-time feedback with translations.

💡 Inspiration

We wanted to create a more immersive and engaging way to learn a new language. Apps like Duolingo are great, but they're often limited to tapping words or filling in blanks. Real-world fluency comes from actual conversation. So we built TalkLingua, where you can speak to an AI in a foreign language, get instant feedback, and see the English translation in real time. It's like having a personal tutor who's always available—and never judges.

🛠️ What it does

Duolingo-v2 lets you:

🗣️ Speak directly to an AI in Spanish (or any target language).

🔁 Get real-time responses from the AI in the same language.

📝 Receive a live transcript with English translation for every sentence.

You start a conversation, talk into your mic, and the AI listens, understands, and responds in kind—while translating everything so you know exactly what's going on.

⚙️ How we built it

We combined powerful APIs and services to create a seamless voice-first experience:

🎙 Daily: For real-time voice conversations over WebRTC.

🤖 Anthropic Claude 3 (via API): For generating natural Spanish responses and translating user input.

🧠 OpenAI (optional): For prompt tuning and future expansion into voice analysis.

🦾 Tavus: For personalized video generation, turning language exercises into visual conversation companions.

⚡ Bolt: For faster iteration of building frontend.

🧪 Flask + SSE: Real-time Server-Sent Events streaming the AI and user utterances to the frontend.

We also used queueing and threading to handle continuous voice input/output with zero delay, and the anthropic.messages.create tool call system to ensure structured, tutor-style replies with both Spanish and English.

🧠 AI Prompting Highlights

We prompted Claude with:

"Act as a Spanish tutor. Respond in Spanish, then provide an English translation. Use structured tool calls only—no extra reasoning or chit-chat."

This made Claude return clear, concise, structured responses like:

{ "spanish_response": "¿Qué te gustaría comer?",

"english_translation": "What would you like to eat?" }

🖥️ Live Demo

Main Demo

Alt Demo - low sound

🌍 Use Cases

Language learners who want to speak more than they tap.

Travelers prepping for real-life conversations.