Deep-Gabriel is an autonomous multi-agent research pipeline that orchestrates complex, deep-dive web research using state-of-the-art Large Language Models. Leveraging LangGraph and DeepSeek, it decomposes complex user queries into sub-topics, conducts parallel multi-threaded research through specialized agents, and compiles comprehensive, fully-formatted reports.
Deep-Gabriel uses a robust supervisor-worker architecture:
- Scoping Agent: Analyzes user queries and determines if clarification is needed before generating a comprehensive research brief.
- Supervisor Agent: Deconstructs the research brief into discrete topics, delegates them to researcher agents, and coordinates the overall workflow to ensure maximum depth and coverage.
- Researcher Agents: Specialized agents that use tool calling to navigate the web, summarize documents, and synthesize insights without hitting rate limits or context window restrictions.
- Report Generation: Aggregates all insights into a polished, downloadable DOCX document, properly formatted with native LaTeX math and markdown support.
- Parallel Execution: Conducts independent research threads simultaneously.
- Automated DocX Generation: Produces well-structured, print-ready documents out-of-the-box.
- DeepSeek Integration: Uses the highly capable DeepSeek language models to evaluate search results and draw conclusions.
- Gradio Interface: Provides a clean, modern web interface for users to enter prompts and download their final papers.
- Ensure dependencies are installed and the environment is configured.
- Provide your
DEEPSEEK_API_KEYin the.envfile. - Run the application:
python main.py
- Enter your research topic in the interface and click Generate Research Paper.
- LangChain & LangGraph
- DeepSeek
- Gradio
- python-docx
You can find an example of the generated research paper in the outputs/ directory. This demonstrates the depth, structuring, and docx formatting capabilities of the multi-agent system.
We welcome contributions to improve Deep-Gabriel! Next steps for the project could include:
- Better Math Rendering: Integrating a robust LaTeX renderer to properly compile math symbols natively within the
.docxoutput files.