A streamlined research system that generates comprehensive Wikipedia-style articles through multi-perspective expert engagement and focused topic exploration.
BREEZE is inspired by STORM (Synthesis of Topic Outlines through Retrieval and Multi-perspective Question Asking) developed by Shao et al. While STORM focuses on broad research capabilities, BREEZE refines this approach specifically for Wikipedia-style article generation with:
- Streamlined architecture focused on article generation
- Enhanced topic validation and scoping
- Structured expert interview process
- Robust citation handling and fact-checking
-
Balanced Multi-Perspective Research:
- Simulates conversations between diverse subject matter experts
- Ensures comprehensive coverage of different viewpoints
- Maintains neutrality in topic exploration
-
Expert Interview System:
- Conducts focused interviews with AI experts
- Gathers detailed information and citations
- Validates information through cross-referencing
-
Structured Article Generation:
- Creates well-organized articles with:
- Clear section outlines
- Proper citations and references
- Consistent writing style
- Wikipedia-style formatting
- Creates well-organized articles with:
-
Zonal Topic Exploration:
- Efficiently scopes and defines research boundaries
- Maintains focus on relevant subject areas
- Ensures appropriate depth of coverage
-
Topic Input and Validation
- Submit your research topic
- System validates and scopes the subject area
- Establishes clear research boundaries
-
Research and Synthesis
- Generates structured outline
- Creates expert personas for different perspectives
- Conducts targeted expert interviews
- Refines outline based on gathered insights
-
Article Generation
- Writes section drafts
- Integrates expert insights
- Adds proper citations
- Delivers polished final article
-
Installation
pip install breeze-agent
-
API Keys Setup
BREEZE requires the following API keys to be set as environment variables:
ANTHROPIC_API_KEY: For outline generations, expert interview interactions and section writingOPENAI_API_KEY: For embedding model accessTAVILY_API_KEY: For web search capabilities
You can set these using environment variables:
export ANTHROPIC_API_KEY="your-key-here" export OPENAI_API_KEY="your-key-here" export TAVILY_API_KEY="your-key-here"
-
Basic Usage
from web_research_graph.graph import graph # Initialize research query result = await graph.ainvoke({ "messages": "Your research topic here" }) # Display results using markdown from IPython.display import Markdown, display display(Markdown(result["article"]))
-
Output The system will generate a well-structured Wikipedia-style article based on your research topic, complete with citations and multiple expert perspectives.
- Technical: "Impact of Large Language Models on Software Development"
- Business: "The Rise of AI-Powered Customer Service"
- General: "History and Evolution of Electric Vehicles"
- Quality depends on available online sources
- May require topic refinement for very broad subjects
- Citations limited to publicly accessible sources
BREEZE builds upon the innovative foundation laid by STORM (Shao et al.), as documented in the LangGraph documentation. We've refined their approach of outline-driven research and multi-perspective conversations while adding specialized enhancements for Wikipedia-style article generation.