README_AI_Talk.md

# AI Friends Talk

## Real AI Model Integration

This project implements AI Friends Talk with **REAL** AI model API calls, following the structure from `day1.ipynb`.

### Two Versions Available:

#### 1. AI_Talk.py (Streamlit - REAL AI APIs) ✅
- **Uses REAL API calls** to actual AI models
- **Alex**: Powered by Groq llama3-70b-8192 (Witty debater)
- **Blake**: Powered by Gemini 2.0 Flash (Creative optimist) 
- **Charlie**: Powered by Gemini 1.5 Flash (Logical analyst)
- Interactive Streamlit interface with multilingual support
- Run with: `streamlit run AI_Talk.py`

#### 2. ai-friends-talk.html (HTML - Simulated Responses)
- Uses simulated responses for demonstration
- Good for UI/UX testing without API costs
- Compatible with the multilingual JavaScript system
- Open directly in browser

### API Setup Required for Real AI Integration:

Create a `.env` file with:
```
GROQ_API_KEY=your_groq_api_key_here
GOOGLE_API_KEY=your_google_api_key_here
```

### Character Personalities (Real AI Models):

1. **Alex (Groq llama3-70b-8192)**
   - Witty and charismatic AI debater
   - Sharp sense of humor, plays devil's advocate
   - Uses vivid examples and thought-provoking questions

2. **Blake (Gemini 2.0 Flash)** 
   - Imaginative and boundlessly optimistic
   - Natural storyteller with beautiful metaphors
   - Finds connections between unrelated concepts

3. **Charlie (Gemini 1.5 Flash)**
   - Thoughtful and systematic analyst
   - Scientific curiosity and methodical thinking
   - Breaks down complex ideas logically

### Test Results:
✅ All three AI models tested and working correctly
✅ Real API calls to Groq and Google Gemini
✅ Character personalities distinct and engaging
✅ Multilingual support (English, Vietnamese, German)

### Usage:
1. Install dependencies: `pip install streamlit openai anthropic groq python-dotenv`
2. Set up your API keys in `.env` file
3. Run: `streamlit run AI_Talk.py`
4. Select a topic and watch real AI models debate!

---

**Note**: The Streamlit version (AI_Talk.py) uses REAL AI model APIs as requested. The HTML version is for demonstration purposes with simulated responses.