AI Blog Writer Agent

An AI-powered blog writing system that uses web research (Tavily API), RAG (Retrieval-Augmented Generation) with LangChain, and OpenRouter for flexible LLM access to generate high-quality blog posts.

Features

• 🔍 Web Research: Automatically researches topics using Tavily API
• 📚 RAG System: Uses LangChain and FAISS for context-aware generation
• 🤖 Flexible AI Models: Access multiple LLMs via OpenRouter (including free models)
• 📝 Quality Content: Generates 1000-1500 word blog posts with proper structure
• 📖 Source Citations: Automatically includes source references
• 🔍 Observability: Built-in LangSmith integration for tracing and debugging

Prerequisites

• Python 3.10 or higher
• OpenRouter API key (get it at https://openrouter.ai/)
• Tavily API key (get it at https://tavily.com/)

Installation

1. Navigate to the project directory:

   cd "/media/mats/3c24094c-800b-4576-a390-d23a6d7a02291/workspace/test_ai_gen/blog writter"

2. Activate the virtual environment:

   source venv/bin/activate

3. Create a .env file in the project root:

   nano .env

4. Add your API keys to the .env file:

   OPENROUTER_API_KEY=your_openrouter_key_here
   TAVILY_API_KEY=your_tavily_key_here
   
   # Optional: LangSmith for observability and debugging
   LANGCHAIN_TRACING_V2=true
   LANGCHAIN_API_KEY=your_langsmith_api_key_here
   LANGCHAIN_PROJECT=blog-writer-agent
   

Usage

Basic Usage

1. Activate the virtual environment:

   source venv/bin/activate

2. Edit the topic in main.py:

   if __name__ == "__main__":
       generate_blog("Your Topic Here")

3. Run the script:

   python main.py

4. Find your blog post in the output/ directory

Changing the AI Model

Edit config.py to change the OpenRouter model:

# Free models (recommended for testing)
OPENROUTER_MODEL = "meta-llama/llama-3-70b-instruct"
# OPENROUTER_MODEL = "google/gemma-7b-it:free"
# OPENROUTER_MODEL = "mistralai/mistral-7b-instruct:free"

# Paid models (better quality)
# OPENROUTER_MODEL = "anthropic/claude-3.5-sonnet"
# OPENROUTER_MODEL = "openai/gpt-4-turbo"

Project Structure

blog-writer-agent/
├── venv/                    # Virtual environment
├── .env                     # API keys (create this)
├── .gitignore              
├── requirements.txt         # Dependencies
├── config.py               # Configuration
├── main.py                 # Main script
├── agents/
│   ├── __init__.py
│   ├── researcher.py       # Tavily web search agent
│   ├── rag_manager.py      # RAG system with FAISS
│   └── writer.py           # Blog post generation agent
└── output/                 # Generated blog posts

How It Works

1. Research Phase: The Researcher Agent searches the web using Tavily API to gather relevant, up-to-date information about your topic.

2. RAG Building: The RAG Manager processes the research data, splits it into chunks, and stores it in an in-memory FAISS vector store with embeddings.

3. Context Retrieval: When generating content, the system retrieves the most relevant chunks from the vector store.

4. Content Generation: The Writer Agent uses an OpenRouter LLM with the retrieved context to generate a comprehensive, well-structured blog post.

5. Output: The final blog post is saved as a markdown file with source citations.

Example Output

output/blog_The_Future_of_AI_in_Healthcare.md

The output includes:

• Properly formatted markdown
• Clear section headings
• 1000-1500 words of content
• Source citations at the end

Customization

Adjust Research Depth

Edit main.py to change the number of sources:

research_data = researcher.search(
    f"Latest information about {topic}",
    max_results=10  # Increase for more sources
)

Change Writing Style

Modify the style parameter in main.py:

blog_post = writer.generate_blog_post(
    topic=topic,
    context_docs=context_docs,
    style="technical"  # Options: professional, casual, technical, etc.
)

Adjust RAG Parameters

Edit agents/rag_manager.py:

chunk_size=1000,      # Increase for larger chunks
chunk_overlap=200,    # Adjust overlap

Troubleshooting

"API key not found" error

Make sure you've created a .env file with both API keys.

Import errors

Ensure you've activated the virtual environment:

source venv/bin/activate

"No research data found"

Check your Tavily API key and internet connection.

Model not available

Some OpenRouter models require credits. Check available free models at https://openrouter.ai/models

LangSmith Observability

The system includes built-in LangSmith integration for comprehensive observability:

What You Get

• Complete LLM Interaction Logs: Every prompt and response is automatically logged
• Agent Identification: See which agent (Writer, Planner, Scorer) made each call
• Session Tracking: Group related interactions by session ID
• Performance Metrics: Response times, token usage, and costs
• Error Tracking: Automatic error logging and debugging information
• Visual Interface: Web dashboard at https://smith.langchain.com

Setup (Optional)

1. Sign up for LangSmith (free tier available): https://smith.langchain.com
2. Get your API key from the LangSmith dashboard
3. Add to your .env file:

   LANGCHAIN_TRACING_V2=true
   LANGCHAIN_API_KEY=your_langsmith_api_key_here
   LANGCHAIN_PROJECT=blog-writer-agent
   

Viewing Traces

• After running the blog generator, check the console output for session information
• Visit https://smith.langchain.com to view detailed traces
• Filter by session ID to see all interactions for a specific run
• Analyze performance, debug issues, and optimize prompts

Cost Estimates

• Tavily API: Free tier includes 1,000 searches/month
• OpenRouter:
  • Free models: $0
  • Paid models: ~$0.01-0.10 per blog post (depending on model)
• LangSmith: Free tier includes 1,000 traces/month

License

MIT License - Feel free to modify and use as needed.

Enter to venv

source venv/bin/activate