LinkedIn Icebreaker Bot

An AI-powered assistant that generates personalized icebreakers and conversation starters based on LinkedIn profiles. Built with IBM watsonx.ai and LlamaIndex, it helps make professional introductions more personal and engaging.

🌟 Features

• LinkedIn Profile Analysis: Extract professional data using ProxyCurl API or use mock data
• AI-Powered Insights: Generate interesting facts about a person's career/education
• Personalized Q&A: Answer specific questions about the person's background
• Two Interfaces: Command-line tool for quick usage and web UI for user-friendly interaction
• Flexible: Use mock data for practice or connect to real LinkedIn profiles

🚀 Quick Start

Prerequisites

• Python 3.11+, < 3.13
• A ProxyCurl API key (optional - mock data available)

Installation

1. Clone the repository:

git clone https://github.com/HaileyTQuach/icebreaker.git
cd icebreaker

2. Create a virtual environment:

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies:

pip install -r requirements.txt

4. (Optional) Add your ProxyCurl API key to config.py:

PROXYCURL_API_KEY = "your-api-key-here"

Using the Command Line Interface

Run the bot using the terminal:

# Use mock data (no API key needed)
python main.py --mock

# OR use a real LinkedIn profile
python main.py --url "https://www.linkedin.com/in/username/" --api-key "your-api-key"

Using the Web Interface

Launch the web app:

python app.py

Then open your browser to the URL shown in the terminal (typically http://127.0.0.1:7860).

🧠 How It Works

The Icebreaker Bot uses a Retrieval-Augmented Generation (RAG) pipeline:

1. Data Extraction: LinkedIn profile data is retrieved via ProxyCurl API or mock data
2. Text Processing: Profile data is split into manageable chunks
3. Vector Embedding: Text chunks are converted to vector embeddings using IBM watsonx
4. Storage: Embeddings are stored in a vector database
5. Query & Generation: When asked a question, relevant profile sections are retrieved and an IBM watsonx LLM generates contextually accurate responses

🛠️ Project Structure

icebreaker_bot/
├── requirements.txt           # Dependencies
├── config.py                  # Configuration settings
├── modules/
│   ├── __init__.py
│   ├── data_extraction.py     # LinkedIn profile data extraction
│   ├── data_processing.py     # Data splitting and indexing
│   ├── llm_interface.py       # LLM setup and interaction
│   └── query_engine.py        # Query processing and response generation
├── app.py                     # Gradio web interface
└── main.py                    # CLI application

📝 Examples

Here are some example questions you can ask:

• "What is this person's current job title?"
• "Where did they get their education?"
• "What skills do they have related to machine learning?"
• "How long have they been working at their current company?"
• "What was their career progression?"

🧪 Customization

Using Different LLM Models

You can switch between available models:

python main.py --mock --model "meta-llama/llama-3-3-70b-instruct"

Or in the web interface, select from the dropdown menu.

Adjusting Response Style

Edit the prompt templates in config.py to change how responses are generated:

INITIAL_FACTS_TEMPLATE = """
You are an AI assistant that provides detailed answers based on the provided context.
...
"""

👩‍💻 Development

For Beginners

If you're learning to build this project from scratch, check out the 1-start branch:

git checkout 1-start

This branch contains starter files with TODOs and guidance for implementation.

Running Tests

Test individual components:

# Test data extraction
python -c "from modules.data_extraction import extract_linkedin_profile; print(extract_linkedin_profile('https://www.linkedin.com/in/username/', mock=True))"

# Test the entire pipeline
python main.py --mock --test

📄 License

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

🙏 Acknowledgments

• IBM watsonx.ai for providing the LLM and embedding models
• LlamaIndex for the data indexing and retrieval framework
• ProxyCurl for LinkedIn profile data extraction
• Eden Marco for the original tutorial inspiration