Ice Breaker LLM

An AI-powered application that generates personalized conversation starters and insights by analyzing LinkedIn profiles. Simply input a person's name, and the system will find their LinkedIn profile, extract key information, and create engaging ice breakers using local LLM models.

🚀 Features

• Automated LinkedIn Profile Discovery: Uses Tavily search API to find LinkedIn profiles from names
• AI-Powered Profile Analysis: Leverages Ollama (Llama 3/3.1) for intelligent content processing
• Smart Content Generation: Creates personalized summaries, interesting facts, topics of interest, and ice breakers
• Modern Web Interface: Clean, responsive Flask-based web application
• Privacy-Focused: Uses local LLM models (Ollama) instead of cloud-based APIs
• Professional Data Extraction: Integrates with Proxycurl API for comprehensive LinkedIn data

🛠️ Technology Stack

• Backend: Python, Flask
• AI/ML: LangChain, Ollama (Llama 3/3.1), Pydantic
• APIs: Proxycurl (LinkedIn data), Tavily (web search)
• Frontend: HTML, CSS, JavaScript
• Environment: Python-dotenv for configuration

📋 Prerequisites

Before running this application, ensure you have:

1. Python 3.8+ installed
2. Ollama installed and running locally
3. API Keys:
   • Proxycurl API key (for LinkedIn data extraction)
   • Tavily API key (for web search)

🔧 Installation

1. Clone the repository:

   git clone https://github.com/DahalRojan/ice-breaker-llm.git
   cd ice-breaker-llm

2. Install dependencies:

   pip install -r requirements.txt

3. Set up Ollama:

   # Install Ollama (visit https://ollama.ai for installation instructions)
   # Pull required models
   ollama pull llama3
   ollama pull llama3.1

4. Configure environment variables: Create a .env file in the project root:

   PROXYCURL_API_KEY=your_proxycurl_api_key_here
   TAVILY_API_KEY=your_tavily_api_key_here

🚀 Usage

1. Start the application:

   python app.py

2. Access the web interface: Open your browser and navigate to http://localhost:5000

3. Generate ice breakers:

   • Enter a person's full name in the search form
   • Click "Do Your Magic"
   • Wait for the AI to process the information
   • View the generated summary, facts, interests, and ice breakers

🏗️ System Architecture

Workflow

1. Input Processing: User enters a person's name through the web interface
2. Profile Discovery: Tavily search API finds the corresponding LinkedIn profile URL
3. Data Extraction: Proxycurl API retrieves comprehensive LinkedIn profile data
4. AI Analysis: Three specialized LangChain pipelines process the data:
   • Summary Chain: Creates professional summary and interesting facts
   • Interests Chain: Identifies topics that might interest the person
   • Ice Breaker Chain: Generates personalized conversation starters
5. Output Generation: Results are formatted and displayed with profile picture

Key Components

• app.py: Flask web server and API endpoints
• ice_breaker.py: Core orchestration logic
• agents/: LinkedIn profile lookup agents using LangChain
• chains/: LLM processing chains for different output types
• third_parties/: External API integrations (LinkedIn, Twitter)
• output_parsers.py: Pydantic models for structured data output
• tools/: Utility functions for web search and data processing

🔧 Configuration Options

Environment Variables

Create a .env file with the following variables:

# Required API Keys
PROXYCURL_API_KEY=your_proxycurl_api_key_here
TAVILY_API_KEY=your_tavily_api_key_here

# Optional: OpenAI API (if you want to use GPT instead of Ollama)
# OPENAI_API_KEY=your_openai_api_key_here

Mock Mode

For testing purposes, you can use mock LinkedIn data by modifying third_parties/linkedin.py:

# In ice_breaker.py, change:
linkedin_data = scrape_linkedin_profile(linkedin_profile_url=linkedin_username, mock=True)

📡 API Endpoints

POST /process

Processes a person's name and returns ice breaker information.

Request Body:

{
  "name": "John Doe"
}

Response:

{
  "summary_and_facts": {
    "summary": "Professional summary...",
    "facts": ["Fact 1", "Fact 2"]
  },
  "interests": {
    "topics_of_interest": ["Topic 1", "Topic 2", "Topic 3"]
  },
  "ice_breakers": {
    "ice_breakers": ["Ice breaker 1", "Ice breaker 2"]
  },
  "picture_url": "https://profile-pic-url.com"
}

🧪 Testing

To test the application with mock data:

1. Set mock=True in the scrape_linkedin_profile function call
2. The system will use predefined mock data instead of making API calls
3. This is useful for development and testing without consuming API credits

🤝 Contributing

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

📄 License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

🛠️ Troubleshooting

Common Issues

1. Ollama not running: Ensure Ollama is installed and running (ollama serve)
2. Missing API keys: Check your .env file for correct API key configuration
3. Model not found: Pull required models (ollama pull llama3 and ollama pull llama3.1)
4. Port already in use: Change the port in app.py if 5000 is occupied

Support

For issues and questions, please open an issue on the GitHub repository.

📸 Screenshots