Multi-Agent Brainstorming System

中文版 / Chinese Version

Try it now: https://lum1104.github.io/agent-brainstorm/

Example of Brainstorm: Idea inspired by Emotion-LLaMA

A Python-based AI agent system that utilizes Google's Gemini API to facilitate structured brainstorming sessions through multiple AI personas. The system runs as a command-line application with an interactive workflow.

Overview

This system implements a 5-stage brainstorming methodology using AI agents to generate, evaluate, and refine ideas for both project development and research papers. The entire process runs locally with real-time web search integration and ArXiv research capabilities.

graph TD;
        __start__([<p>__start__</p>]):::first
        ask_for_pdf_path(ask_for_pdf_path)
        process_pdf(process_pdf)
        context_generation(context_generation)
        persona_generation(persona_generation)
        divergent_ideation(divergent_ideation)
        collaborative_discussion(collaborative_discussion)
        user_filter_ideas(user_filter_ideas)
        red_team_critique(red_team_critique)
        convergent_evaluation(convergent_evaluation)
        user_select_idea(user_select_idea)
        ask_for_arxiv_search(ask_for_arxiv_search)
        arxiv_search(arxiv_search)
        implementation_planning(implementation_planning)
        user_feedback_on_plan(user_feedback_on_plan)
        __end__([<p>__end__</p>]):::last
        __start__ --> ask_for_pdf_path;
        arxiv_search --> implementation_planning;
        ask_for_arxiv_search -.-> arxiv_search;
        ask_for_arxiv_search -.-> implementation_planning;
        ask_for_pdf_path -.-> context_generation;
        ask_for_pdf_path -.-> process_pdf;
        collaborative_discussion --> user_filter_ideas;
        context_generation --> persona_generation;
        convergent_evaluation --> user_select_idea;
        divergent_ideation --> collaborative_discussion;
        implementation_planning --> user_feedback_on_plan;
        persona_generation --> divergent_ideation;
        process_pdf --> context_generation;
        red_team_critique --> convergent_evaluation;
        user_feedback_on_plan -. &nbsp;END&nbsp; .-> __end__;
        user_feedback_on_plan -.-> user_select_idea;
        user_filter_ideas --> red_team_critique;
        user_select_idea --> ask_for_arxiv_search;
        classDef default fill:#f2f0ff,line-height:1.2
        classDef first fill-opacity:0
        classDef last fill:#bfb6fc

Loading

Getting Started

Installation

git clone https://github.com/Lum1104/agent-brainstorm.git
cd agent-brainstorm

conda create -n brainstorm python=3.12
conda activate brainstorm

pip install -r requirements.txt
python main.py

API Key Setup

1. Visit Google AI Studio
2. Create a free account if needed
3. Generate an API key (for FREE)
4. Set the API key as an environment variable:

   export GOOGLE_API_KEY="your-api-key-here"

   Or the application will prompt you to enter it when running.

Project Structure

agent-brainstorm/
├── main.py                     # Main entry point
├── brainstorm_tool/
│   ├── agents/
│   │   ├── workflow.py         # Core brainstorming workflow
│   │   ├── schemas.py          # Data validation schemas
│   │   └── prompts.py          # Agent prompt templates
│   └── utils/
│       ├── ui.py               # User interface utilities
│       └── file_utils.py       # File processing utilities
├── requirements.txt            # Python dependencies
└── README.md

Features

🤖 Multi-Agent Architecture

• Preview Agent: Assembles expert teams with distinct personas
• RAG Agent: Provides contextual research and background information using web search and ArXiv
• Ideation Agents: Generate diverse ideas from different perspectives
• Critic Agent: Evaluates and ranks generated ideas with red team critique
• Expert Agent: Creates detailed implementation plans

📋 Two Brainstorming Modes

1. Project Ideas: For products, features, and development projects
2. Research Papers: For academic topics, studies, and research questions

🔄 6-Stage Workflow

1. Context Generation: Gather relevant background information via web search
2. Define Direction & Assemble Team: Configure personas based on your topic
3. Divergent Ideation: Generate multiple ideas from different perspectives
4. Red Team Critique: Challenge ideas with devil's advocate analysis
5. Convergent Evaluation: Analyze, cluster, and rank ideas
6. Final Document Generation: Create detailed plans for selected ideas

How to Use

Step-by-Step Process

Step 1: Configuration

• Select your brainstorming type (Project or Research Paper)
• Enter your topic (e.g., "Using LLM for the course project of HCI")
• Optionally provide a PDF file for additional context

Step 2: Context Generation

The system automatically searches the web for relevant information about your topic.

Step 3: Expert Team Assembly

The Preview Agent assembles a team of 4 expert personas relevant to your topic, each with:

• Role: Their area of expertise
• Goal: What they aim to achieve
• Backstory: Their professional background

Step 4: Divergent Ideation

Each expert persona generates 5 unique ideas. You can:

• Review all generated ideas
• Filter out ideas you want to exclude
• Select the most promising concepts

Step 5: Red Team Critique

A devil's advocate agent challenges each idea with critical analysis.

Step 6: Convergent Evaluation

The Critic Agent:

• Analyzes and clusters similar ideas
• Evaluates each concept on multiple criteria
• Provides detailed justifications incorporating red team feedback
• Ranks the top 3 recommendations

Step 7: Implementation Planning

Select one of the top ideas to generate:

• For Projects: Implementation plan with requirements, tech stack, timeline, and resources
• For Research: Research outline with methodology, literature review, and publication targets

Output

The system generates a comprehensive Markdown report containing:

• Complete session transcript
• All generated ideas and critiques
• Final evaluation and ranking
• Detailed implementation plan
• Relevant academic references from ArXiv

Example Use Cases

Project Development

• "Mobile app for university students"
• "AI-powered learning platform"
• "Sustainable energy management system"

Research Papers

• "Machine learning in healthcare"
• "Human-computer interaction studies"
• "Climate change mitigation strategies"