🎭 Conflict Conversation Generator

A Python project that simulates realistic conflict conversations between two virtual agents with configurable personality traits. Uses RAG (Retrieval-Augmented Generation) to create contextually relevant dialogues based on personality examples.

✨ Features

• 🎭 Multiple Personality Types: Aggressive, defensive, passive, open-minded, emotional, logical
• 🔍 RAG System: Retrieves relevant personality examples using embeddings
• 🤖 Dual Generation: Local LLM (Mistral-7B/Zephyr) + OpenAI integration
• 💬 Topic-Specific: Realistic conflicts about household chores, money, parenting, etc.
• 🔄 Interactive CLI: Easy-to-use command-line interface
• 📊 ARM Compatible: Works on Apple Silicon Macs without FAISS issues
• 🆓 Free & Open Source: No paid APIs required (OpenAI optional)

🚀 Quick Start

1. Installation

# Clone the repository
git clone <your-repo-url>
cd Conflict-Simulation-Lab

# Install backend dependencies
pip install -r backend/requirements.txt

2. Basic Usage

# Interactive mode (recommended)
python run_rag_system.py --interactive

# Command line mode
python run_rag_system.py --personality-a aggressive --personality-b defensive --topic "money and financial decisions"

# List available personalities
python run_rag_system.py --list-personalities

3. OpenAI Enhancement (Optional)

For more realistic, topic-specific conversations:

# Set your OpenAI API key
export OPENAI_API_KEY="your-api-key-here"

# Use OpenAI for generation
python run_rag_system.py --interactive --use-openai

# Or specify directly
python run_rag_system.py --personality-a aggressive --personality-b defensive --topic "parenting styles" --use-openai

🎯 How It Works

Current System (Local LLM)

1. Embedding: Converts personality examples to vectors
2. Retrieval: Finds most relevant examples for each personality
3. Generation: Local LLM creates dialogue using retrieved examples
4. Result: Personality-styled conversation (may be generic)

OpenAI Enhanced System

1. Topic Analysis: Understands specific conflict scenarios
2. Context Integration: Combines personality + topic context
3. Realistic Generation: Creates topic-specific arguments
4. Result: Realistic, topic-aware conflict conversations

📊 Comparison: Local vs OpenAI

Aspect	Local LLM	OpenAI
Topic Specificity	Generic conflicts	Realistic, topic-aware
Cost	Free	~$0.002 per conversation
Speed	Slower (local processing)	Fast (API)
Quality	Good personality style	Excellent realism
Privacy	100% local	Data sent to OpenAI

🎭 Available Personalities

• Aggressive: Confrontational, direct, forceful
• Defensive: Protective, justifying, reactive
• Passive: Yielding, non-confrontational, submissive
• Open-minded: Receptive, compromising, understanding
• Emotional: Feeling-driven, expressive, sensitive
• Logical: Analytical, fact-based, systematic

💬 Conversation Topics

• Household chores and responsibilities
• Money and financial decisions
• Parenting styles and discipline
• Work-life balance
• Social plans and commitments
• Custom topics

🛠️ Project Structure

Conflict-Simulation-Lab/
├── run_rag_system.py          # Main runner script
├── backend/
│   ├── main.py               # Flask API server
│   ├── requirements.txt      # Backend dependencies
│   └── models/               # AI model implementations
│       ├── __init__.py       # Package initialization
│       ├── rag_consumer_complaints.py  # Main RAG system
│       ├── compare_enhanced_rag.py     # Comparison script
│       ├── requirements_enhanced_rag.txt # Model dependencies
│       └── ENHANCED_RAG_README.md      # Model documentation
├── frontend/                 # React frontend application
├── data/
│   ├── personality_index.pkl  # Embedding index
│   └── style_examples.json    # Personality examples
└── README.md                 # This file

🔧 Advanced Usage

Testing Different Systems

Compare local vs OpenAI generation

python test_openai_comparison.py

Test topic specificity

python test_openai_comparison.py

Custom Personalities

Edit data/style_examples.json to add new personality types:

{
  "sarcastic": {
    "description": "Uses irony and wit in communication",
    "tone": "Ironic, witty, sometimes cutting",
    "examples": [
      "Oh, great. Another brilliant idea.",
      "Well, that went exactly as planned...",
      "I'm shocked, shocked I tell you."
    ]
  }
}

LangChain Integration

For advanced features like conversation memory and chaining:

Install LangChain version

pip install -r requirements_langchain.txt

Run LangChain version

python langchain_integration.py

🎯 RAG System Explanation

The system uses Retrieval-Augmented Generation to improve dialogue quality:

1. Retrieval: Finds relevant personality examples from the database
2. Augmentation: Combines examples with the generation prompt
3. Generation: Creates dialogue that matches the retrieved style

Why it works better:

• Provides concrete examples of how each personality communicates
• Ensures consistency in personality expression
• Improves topic relevance when combined with OpenAI

🚀 Strengthening the RAG System

1. Expand the Database

• Add more personality examples
• Include topic-specific examples
• Add conversation context examples

2. Improve Retrieval

• Use better embedding models
• Implement reranking
• Add semantic search

3. Enhanced Generation

• Use larger models (GPT-4, Claude)
• Implement conversation memory
• Add multi-turn dialogue

🔑 OpenAI Setup

1. Get API Key: Visit OpenAI Platform
2. Set Environment Variable:

   export OPENAI_API_KEY="your-key-here"

3. Test Integration:

   python test_openai_comparison.py

🐛 Troubleshooting

Common Issues

Segmentation Fault on ARM Mac:

• The system is designed to work without FAISS
• Uses pure numpy-based retrieval
• Should work on Apple Silicon without issues

OpenAI API Errors:

• Check your API key is valid
• Ensure you have sufficient credits
• Verify internet connection

Local Generation Issues:

• Check model downloads
• Verify sufficient RAM (8GB+ recommended)
• Try different models in generate.py

Performance Tips

• Local: Use GPU if available for faster generation
• OpenAI: Batch requests to reduce API calls
• Memory: Close other applications for local generation

🤝 Contributing

1. Fork the repository
2. Create a feature branch
3. Add your improvements
4. Test thoroughly
5. Submit a pull request

📄 License

This project is open source and available under the MIT License.

🙏 Acknowledgments

• Hugging Face for open-source models
• OpenAI for enhanced generation capabilities
• SentenceTransformers for embeddings
• The open-source community for tools and libraries