AI Blog Writer Agent

An AI-powered technical blog writing agent built with LangGraph, LangChain, and Streamlit. The agent can research topics, plan blog content, generate sections, and even create images for blog posts.

Features

• AI-Powered Research: Automatically researches topics using Tavily search
• Smart Planning: Creates detailed blog outlines with tasks, goals, and word targets
• Section Generation: Generates individual blog sections using LLM
• Image Generation: Creates relevant images using Gemini
• Web Interface: User-friendly Streamlit UI for managing the blog writing process
• Export Options: Download blog posts as Markdown or bundled with images

Architecture

The project consists of two main files:

• bwa_backend.py: LangGraph-based backend that handles:

  • Routing (determines if research is needed)
  • Research (fetches relevant information)
  • Orchestration (creates blog plan)
  • Worker nodes (generate content sections)
  • Reducer (merges content and handles images)

• bwa_frontend.py: Streamlit web interface with:

  • Topic input and date selection
  • Real-time progress tracking
  • Plan, Evidence, Preview, Images, and Logs tabs
  • Download functionality

Installation

1. Clone the repository and navigate to the project directory:

bash
cd AI-blog-agent

2. Install the required dependencies:

bash
pip install streamlit langgraph langchain-openai python-dotenv pydantic pandas

3. Set up environment variables in a .env file:

env
# Required for LLM calls
OPENAI_API_KEY=your_openai_api_key

# Optional - for research functionality
TAVILY_API_KEY=your_tavily_api_key

# Optional - for image generation
GOOGLE_API_KEY=your_google_api_key

Usage

1. Start the Streamlit application:

bash
streamlit run bwa_frontend.py

2. Open your browser and navigate to http://localhost:8501

3. Enter a topic in the text area and click "🚀 Generate Blog"

4. View results in the different tabs:

   • Plan: Shows the generated blog outline
   • Evidence: Displays research findings
   • Markdown Preview: Shows the generated blog content
   • Images: Displays generated images
   • Logs: Shows execution logs

5. Download the blog as Markdown or as a bundle with images

Blog Generation Modes

The agent supports three modes:

• closed_book: Evergreen content without research
• hybrid: Evergreen content with some research for up-to-date examples
• open_book: News/roundup content requiring recent research

Project Structure

AI-blog-agent/
├── bwa_backend.py          # LangGraph backend
├── bwa_frontend.py        # Streamlit frontend
├── README.md              # This file
├── 1_bwa_basic.ipynb      # Basic blog writing agent notebook
├── 2_bwa_improved_prompting.ipynb  # Improved prompting techniques
├── 3_bwa_research.ipynb   # Research integration notebook
├── 4_bwa_research_fine_tuned.ipynb # Fine-tuned research workflow
├── 5_bwa_image.ipynb      # Image generation capabilities
├── tavily_test.ipynb      # Tavily search testing
├── *.md                   # Generated blog posts
└── images/                # Generated images directory

Requirements

• Python 3.10+
• OpenAI API key (for LLM)
• Tavily API key (optional, for research)
• Google API key (optional, for image generation)