monju 0.1.3

A python library for brainstorming by multiple LLMs.

monju: Multi-AI Brainstorming Framework

Monju is a powerful multi-AI brainstorming framework designed to generate, organize, and evaluate ideas using various Large Language Models (LLMs).

The Japanese proverb “三人寄れば文殊の知恵” (San nin yoreba monju no chie) is often translated into English as “Two heads are better than one”. However, a more literal translation would be “When three people gather, they have the wisdom of Manjushri”.

Features

• Generate ideas based on a given theme
• Organize ideas into mindmaps and class diagrams
• Evaluate generated ideas
• Support for multiple LLM providers (OpenAI, Anthropic, Google)
• Customizable parameters for idea generation

The project started as a web app monju.ai, which was released in July 2024.

This library is the core of monju.ai that does not include frontend part.

For general information of monju.ai, see my article from Medium "monju.ai: a creativity enhancement tool that uses multiple generative AIs to generate ideas and organize mind maps".

Brainstorming Process by monju

The brainstorming process by monju consists of three main steps:

1. Generate ideas in list
2. Organize ideas in mindmap and class diagram
3. Evaluate ideas including overall evaluation, good points, and areas for improvement

Each step is implemented as a method in the Monju class.

For the general approach to brainstorming, see Brainstorming - Wikipedia.

Both mindmap and class diagram are described in mermaid format. Output can be drawn by any mermaid-compatible tools mainly in JavaScript, e.g. Mermaid Live Editor.

KJ method (Affinity Diagram)

In Japan, another popular ideation tool "KJ method" is often used as part of brainstorming. This method might be known as "affinity diagram" in English.

The KJ method is to categorize ideas into groups based on their similarity. Each idea is written on a single card, then categorized into groups with labels.

People in Japan hesitate open discussion. Writing in cards and categorizing them into groups might help open mind.

In monju, draw affinity wall diagram in class diagram. This process is called "pseudo KJ method" in this library.

For details of KJ method, see Affinity diagram - Wikipedia.

Installation

pip install monju

Requirements

• Python 3.10 or later
• API keys for the LLM providers

This library uses LLMMaster, a unified interface to access major LLM providers. Set API keys following the instructions from LLM Master, another repository by the same author.

Usage

Here's a basic example of how to use monju:

Batch Execution

import json

from pathlib import Path
from monju import Monju

API_KEY = Path('your_api_keys.txt').read_text(encoding='utf-8')

# Use this function to arrange parameters in dictionary format.
def pack_parameters(**kwargs):
    return kwargs


# Initialize monju with your API keys and brainstorming parameters
params = pack_parameters(
    theme="How to survive in the era of emerging AI?",
    ideas=5,
    freedom=0.2,
    language="EN"
)
bs = Monju(api_keys=API_KEY, verbose=True, **params)

# Start the brainstorming process
bs.brainstorm()

# Show the results
print(f'Result:\n{json.dumps(bs.record, indent=2, ensure_ascii=False)}')

Step-by-Step Execution

There is another way to use monju by calling each method step by step.

The following example works equivalently to the batch execution example above.

import json

from pathlib import Path
from monju import Monju

API_KEY = Path('your_api_keys.txt').read_text(encoding='utf-8')

# Use this function to arrange parameters in dictionary format.
def pack_parameters(**kwargs):
    return kwargs


# Initialize monju with your API keys and brainstorming parameters
params = pack_parameters(
    theme="How to survive in the era of emerging AI?",
    ideas=5,
    freedom=0.2,
    language="EN"
)
bs = Monju(api_keys=API_KEY, verbose=True, **params)

# Start the brainstorming process
print(f"Status: {bs.status}")
bs.generate_ideas()
print(f"Status: {bs.status}")
bs.organize_ideas()
print(f"Status: {bs.status}")
bs.evaluate_ideas()
print(f"Status: {bs.status}")
bs.verify()
print(f"Status: {bs.status}")

# Show the results
print(f'Result:\n{json.dumps(bs.record, indent=2, ensure_ascii=False)}')

Types of bs.status are explained in later section.

Input Parameters

There are several parameters to set for Monju class:

• System parameters:
  • api_keys (str) required: API keys for LLMs in LLMMaster manner
  • verbose (bool): print progress for debugging, default is False
• Brainstorming parameters params:
  • theme (str) required: theme or topic of brainstorming, a.k.a. prompt
  • ideas (int): number of ideas to generate by each LLM
  • freedom (float): value of freedom-looking thinking for each LLM, a.k.a. temperature, between 0 and 1
  • language (str): language for output, default is EN and must be followed by two-letter code defined in ISO 639-1

As shown in the examples above, the brainstorming parameters are passed to Monju class as a dictionary. Use pack_parameters to format.

Output Format

Results are given in JSON format. Here is a real example of print(bs.record) given by the example code above.

{
  "input": {
    "theme": "How to survive in the era of emerging AI?",
    "ideas": 5,
    "freedom": 0.2,
    "language": "EN",
    "idea_generation": {
      "openai": {
        "provider": "openai",
        "model": "gpt-4o-mini",
        "prompt": "\nPurpose: Generate 5 ideas on the following \"Theme\" and meeting the \"Conditions\".\n\nConditions:\n1. Propose ideas with free thinking.\n2. Consider the 5W1H when proposing.\n3. Return the results in bullet points as shown in the \"Format\".\n4. The language of the output is in EN of ISO 639-1 format.\n5. Do not include your explanations.\n6. Do not add any unnecessary decorations to the bullet points.\n\nTheme: How to survive in the era of emerging AI?\n\nFormat:\n- Idea\n",
        "temperature": 0.2
      },
      "anthropic": {
        "provider": "anthropic",
        "model": "claude-3-haiku-20240307",
        "prompt": "\nPurpose: Generate 5 ideas on the following \"Theme\" and meeting the \"Conditions\".\n\nConditions:\n1. Propose ideas with free thinking.\n2. Consider the 5W1H when proposing.\n3. Return the results in bullet points as shown in the \"Format\".\n4. The language of the output is in EN of ISO 639-1 format.\n5. Do not include your explanations.\n6. Do not add any unnecessary decorations to the bullet points.\n\nTheme: How to survive in the era of emerging AI?\n\nFormat:\n- Idea\n",
        "temperature": 0.2
      },
      "google": {
        "provider": "google",
        "model": "gemini-1.5-flash",
        "prompt": "\nPurpose: Generate 5 ideas on the following \"Theme\" and meeting the \"Conditions\".\n\nConditions:\n1. Propose ideas with free thinking.\n2. Consider the 5W1H when proposing.\n3. Return the results in bullet points as shown in the \"Format\".\n4. The language of the output is in EN of ISO 639-1 format.\n5. Do not include your explanations.\n6. Do not add any unnecessary decorations to the bullet points.\n\nTheme: How to survive in the era of emerging AI?\n\nFormat:\n- Idea\n",
        "temperature": 0.2
      }
    },
    "mindmap": {
      "provider": "openai",
      "model": "gpt-4o"
    },
    "class_diagram": {
      "provider": "openai",
      "model": "gpt-4o"
    },
    "idea_evaluation": {
      "openai": {
        "provider": "openai",
        "model": "gpt-4o-mini",
        "prompt": "\nPurpose: Evaluate the \"Mindmap\" based on the \"Theme\" and meeting the \"Conditions\".\n\nTheme: How to survive in the era of emerging AI?\n\nConditions:\n1. Write \"Overall Evaluation\", \"Good Points\", and \"Areas for Improvement\" according to the \"Format\".\n2. The language of the output is in EN of ISO 639-1 format.\n3. Write as concisely and focused on key points as possible.\n4. The mindmap is provided as a class diagram.\n5. Do not add any unnecessary decorations to the bullet points.\n\nMindmap:\nclassDiagram\n    class LifelongLearning {\n        +Embrace lifelong learning\n        +Cultivate adaptability\n        +Continuously update skills\n    }\n    \n    class HumanSkills {\n        +Foster creativity\n        +Develop emotional intelligence\n        +Enhance critical thinking\n        +Invest in unique human skills\n    }\n    \n    class AIIntegration {\n        +Collaborate with AI\n        +Develop specialized skills\n        +Focus on human-AI collaboration\n    }\n    \n    class EthicalAI {\n        +Advocate for ethical AI use\n        +Engage in responsible AI development\n        +Promote equitable AI regulations\n    }\n    \n    class DiverseSkillSet {\n        +Build a diverse skill set\n        +Diversify expertise\n    }\n    \n    class AIUnderstanding {\n        +Develop comprehensive understanding of AI\n        +Make informed decisions\n    }\n    \n    class Entrepreneurship {\n        +Explore entrepreneurial opportunities\n        +Leverage AI for innovative solutions\n    }\n    \n    LifelongLearning --> HumanSkills\n    LifelongLearning --> AIIntegration\n    HumanSkills --> AIIntegration\n    AIIntegration --> EthicalAI\n    EthicalAI --> AIUnderstanding\n    DiverseSkillSet --> AIUnderstanding\n    Entrepreneurship --> AIIntegration\n    Entrepreneurship --> EthicalAI\n\nFormat:\n- Overall Evaluation:\n  - Bullet point\n- Good Points:\n  - Bullet point\n- Areas for Improvement:\n  - Bullet point\n",
        "temperature": 0.7
      },
      "anthropic": {
        "provider": "anthropic",
        "model": "claude-3-haiku-20240307",
        "prompt": "\nPurpose: Evaluate the \"Mindmap\" based on the \"Theme\" and meeting the \"Conditions\".\n\nTheme: How to survive in the era of emerging AI?\n\nConditions:\n1. Write \"Overall Evaluation\", \"Good Points\", and \"Areas for Improvement\" according to the \"Format\".\n2. The language of the output is in EN of ISO 639-1 format.\n3. Write as concisely and focused on key points as possible.\n4. The mindmap is provided as a class diagram.\n5. Do not add any unnecessary decorations to the bullet points.\n\nMindmap:\nclassDiagram\n    class LifelongLearning {\n        +Embrace lifelong learning\n        +Cultivate adaptability\n        +Continuously update skills\n    }\n    \n    class HumanSkills {\n        +Foster creativity\n        +Develop emotional intelligence\n        +Enhance critical thinking\n        +Invest in unique human skills\n    }\n    \n    class AIIntegration {\n        +Collaborate with AI\n        +Develop specialized skills\n        +Focus on human-AI collaboration\n    }\n    \n    class EthicalAI {\n        +Advocate for ethical AI use\n        +Engage in responsible AI development\n        +Promote equitable AI regulations\n    }\n    \n    class DiverseSkillSet {\n        +Build a diverse skill set\n        +Diversify expertise\n    }\n    \n    class AIUnderstanding {\n        +Develop comprehensive understanding of AI\n        +Make informed decisions\n    }\n    \n    class Entrepreneurship {\n        +Explore entrepreneurial opportunities\n        +Leverage AI for innovative solutions\n    }\n    \n    LifelongLearning --> HumanSkills\n    LifelongLearning --> AIIntegration\n    HumanSkills --> AIIntegration\n    AIIntegration --> EthicalAI\n    EthicalAI --> AIUnderstanding\n    DiverseSkillSet --> AIUnderstanding\n    Entrepreneurship --> AIIntegration\n    Entrepreneurship --> EthicalAI\n\nFormat:\n- Overall Evaluation:\n  - Bullet point\n- Good Points:\n  - Bullet point\n- Areas for Improvement:\n  - Bullet point\n",
        "temperature": 0.7
      },
      "google": {
        "provider": "google",
        "model": "gemini-1.5-flash",
        "prompt": "\nPurpose: Evaluate the \"Mindmap\" based on the \"Theme\" and meeting the \"Conditions\".\n\nTheme: How to survive in the era of emerging AI?\n\nConditions:\n1. Write \"Overall Evaluation\", \"Good Points\", and \"Areas for Improvement\" according to the \"Format\".\n2. The language of the output is in EN of ISO 639-1 format.\n3. Write as concisely and focused on key points as possible.\n4. The mindmap is provided as a class diagram.\n5. Do not add any unnecessary decorations to the bullet points.\n\nMindmap:\nclassDiagram\n    class LifelongLearning {\n        +Embrace lifelong learning\n        +Cultivate adaptability\n        +Continuously update skills\n    }\n    \n    class HumanSkills {\n        +Foster creativity\n        +Develop emotional intelligence\n        +Enhance critical thinking\n        +Invest in unique human skills\n    }\n    \n    class AIIntegration {\n        +Collaborate with AI\n        +Develop specialized skills\n        +Focus on human-AI collaboration\n    }\n    \n    class EthicalAI {\n        +Advocate for ethical AI use\n        +Engage in responsible AI development\n        +Promote equitable AI regulations\n    }\n    \n    class DiverseSkillSet {\n        +Build a diverse skill set\n        +Diversify expertise\n    }\n    \n    class AIUnderstanding {\n        +Develop comprehensive understanding of AI\n        +Make informed decisions\n    }\n    \n    class Entrepreneurship {\n        +Explore entrepreneurial opportunities\n        +Leverage AI for innovative solutions\n    }\n    \n    LifelongLearning --> HumanSkills\n    LifelongLearning --> AIIntegration\n    HumanSkills --> AIIntegration\n    AIIntegration --> EthicalAI\n    EthicalAI --> AIUnderstanding\n    DiverseSkillSet --> AIUnderstanding\n    Entrepreneurship --> AIIntegration\n    Entrepreneurship --> EthicalAI\n\nFormat:\n- Overall Evaluation:\n  - Bullet point\n- Good Points:\n  - Bullet point\n- Areas for Improvement:\n  - Bullet point\n",
        "temperature": 0.7
      }
    }
  },
  "output": {
    "elapsed_time": [
      5.969,
      7.166,
      6.533
    ],
    "ideas": {
      "openai": "- Embrace lifelong learning: Continuously update skills and knowledge to stay relevant in a rapidly changing job market.\n- Foster creativity and emotional intelligence: Develop skills that AI cannot replicate, such as empathy, creativity, and critical thinking.\n- Collaborate with AI: Learn to work alongside AI tools to enhance productivity and efficiency in various tasks.\n- Advocate for ethical AI use: Engage in discussions and initiatives that promote responsible AI development and implementation.\n- Build a diverse skill set: Diversify expertise across multiple fields to adapt to various roles that may emerge due to AI advancements.",
      "anthropic": "- Develop a comprehensive understanding of AI capabilities and limitations to make informed decisions.\n- Cultivate adaptability and lifelong learning to stay relevant in a rapidly evolving job market.\n- Invest in developing unique human skills that complement AI, such as creativity, emotional intelligence, and critical thinking.\n- Advocate for ethical AI development and implementation to ensure AI benefits humanity.\n- Explore entrepreneurial opportunities in the AI industry to leverage your expertise and contribute to its responsible growth.",
      "google": "- **Idea:** Develop specialized skills that complement AI, such as creativity, emotional intelligence, and critical thinking.\n- **Idea:** Embrace lifelong learning and adapt to the evolving job market by acquiring new skills and knowledge.\n- **Idea:** Focus on building strong human connections and fostering collaboration to leverage the strengths of both humans and AI.\n- **Idea:** Advocate for ethical AI development and regulations to ensure responsible and equitable use of AI technology.\n- **Idea:** Explore entrepreneurial opportunities that leverage AI to create innovative solutions and address societal challenges."
    },
    "mindmap": "mindmap\n    root [How to survive in the era of emerging AI?]\n        -Lifelong Learning and Adaptability-\n            -Embrace lifelong learning: Continuously update skills and knowledge to stay relevant in a rapidly changing job market.-\n            -Cultivate adaptability and lifelong learning to stay relevant in a rapidly evolving job market.-\n            -Embrace lifelong learning and adapt to the evolving job market by acquiring new skills and knowledge.-\n        -Human Skills Development-\n            -Foster creativity and emotional intelligence: Develop skills that AI cannot replicate, such as empathy, creativity, and critical thinking.-\n            -Invest in developing unique human skills that complement AI, such as creativity, emotional intelligence, and critical thinking.-\n            -Develop specialized skills that complement AI, such as creativity, emotional intelligence, and critical thinking.-\n        -Collaboration with AI-\n            -Collaborate with AI: Learn to work alongside AI tools to enhance productivity and efficiency in various tasks.-\n            -Focus on building strong human connections and fostering collaboration to leverage the strengths of both humans and AI.-\n        -Ethical AI Advocacy-\n            -Advocate for ethical AI use: Engage in discussions and initiatives that promote responsible AI development and implementation.-\n            -Advocate for ethical AI development and implementation to ensure AI benefits humanity.-\n            -Advocate for ethical AI development and regulations to ensure responsible and equitable use of AI technology.-\n        -Diverse Skill Set and Entrepreneurship-\n            -Build a diverse skill set: Diversify expertise across multiple fields to adapt to various roles that may emerge due to AI advancements.-\n            -Explore entrepreneurial opportunities in the AI industry to leverage your expertise and contribute to its responsible growth.-\n            -Explore entrepreneurial opportunities that leverage AI to create innovative solutions and address societal challenges.-\n        -Understanding AI-\n            -Develop a comprehensive understanding of AI capabilities and limitations to make informed decisions.-",
    "class_diagram": "classDiagram\n    class LifelongLearning {\n        +Embrace lifelong learning\n        +Cultivate adaptability\n        +Continuously update skills\n    }\n    \n    class HumanSkills {\n        +Foster creativity\n        +Develop emotional intelligence\n        +Enhance critical thinking\n        +Invest in unique human skills\n    }\n    \n    class AIIntegration {\n        +Collaborate with AI\n        +Develop specialized skills\n        +Focus on human-AI collaboration\n    }\n    \n    class EthicalAI {\n        +Advocate for ethical AI use\n        +Engage in responsible AI development\n        +Promote equitable AI regulations\n    }\n    \n    class DiverseSkillSet {\n        +Build a diverse skill set\n        +Diversify expertise\n    }\n    \n    class AIUnderstanding {\n        +Develop comprehensive understanding of AI\n        +Make informed decisions\n    }\n    \n    class Entrepreneurship {\n        +Explore entrepreneurial opportunities\n        +Leverage AI for innovative solutions\n    }\n    \n    LifelongLearning --> HumanSkills\n    LifelongLearning --> AIIntegration\n    HumanSkills --> AIIntegration\n    AIIntegration --> EthicalAI\n    EthicalAI --> AIUnderstanding\n    DiverseSkillSet --> AIUnderstanding\n    Entrepreneurship --> AIIntegration\n    Entrepreneurship --> EthicalAI",
    "evaluation": {
      "openai": "- Overall Evaluation:\n  - The mindmap effectively outlines key strategies for surviving in the era of emerging AI, focusing on skill development and ethical considerations.\n\n- Good Points:\n  - Emphasizes the importance of lifelong learning and adaptability.\n  - Highlights the need for human skills that AI cannot replicate, such as creativity and emotional intelligence.\n  - Addresses the significance of ethical AI practices and regulations.\n  - Encourages collaboration between humans and AI, fostering innovation.\n\n- Areas for Improvement:\n  - Could provide more specific examples or actionable steps for each category.\n  - May benefit from integrating more on the potential risks of AI and how to mitigate them.\n  - Could explore the impact of AI on various industries in greater detail.",
      "anthropic": "EN\n\nOverall Evaluation:\n- The mindmap provides a comprehensive and well-structured approach to surviving in the era of emerging AI, covering key aspects such as lifelong learning, human skills, AI integration, ethical AI, diverse skill sets, and entrepreneurship.\n\nGood Points:\n- The mindmap emphasizes the importance of lifelong learning, adaptability, and continuously updating skills to stay relevant in the changing landscape.\n- It highlights the need to foster human-centric skills, such as creativity, emotional intelligence, and critical thinking, to complement AI capabilities.\n- The mindmap recognizes the significance of collaboration between humans and AI, as well as the development of specialized skills to integrate AI effectively.\n- It addresses the importance of ethical AI development and responsible use, as well as the need for informed decision-making and comprehensive understanding of AI.\n- The mindmap encourages the exploration of entrepreneurial opportunities and the leveraging of AI for innovative solutions.\n\nAreas for Improvement:\n- The mindmap could benefit from more specific strategies or actionable steps within each of the key areas to provide clearer guidance on how to implement the proposed approaches.\n- Incorporating examples or case studies could further enhance the practical applicability of the mindmap.\n- Considering the inclusion of additional factors, such as the importance of lifelong learning for mental well-being or the role of interdisciplinary collaboration in navigating the AI landscape, could strengthen the mindmap's comprehensiveness.",
      "google": "- **Overall Evaluation:** The mindmap provides a comprehensive overview of key strategies for surviving in the era of emerging AI. \n- **Good Points:**\n    - The mindmap effectively categorizes strategies into distinct but interconnected areas: Lifelong Learning, Human Skills, AI Integration, Ethical AI, Diverse Skill Set, AI Understanding, and Entrepreneurship.\n    - It highlights the importance of human skills and ethical considerations in navigating the AI landscape.\n    - The connections between the categories demonstrate the interconnectedness of different approaches.\n- **Areas for Improvement:**\n    -  The mindmap could benefit from more specific examples and actionable steps within each category. \n    -  Adding visual cues (e.g., colors, icons) could improve clarity and engagement."
    }
  }
}

The record object contains the following information:

• input: Input parameters for the brainstorming process.
  • theme: same as the input parameter.
  • ideas: same as the input parameter.
  • freedom: same as the input parameter.
  • language: same as the input parameter.
  • idea_generation: LLM information used for idea generation.
  • mindmap: LLM information used for mindmap generation.
  • class_diagram: LLM information used for class diagram generation.
  • idea_evaluation: LLM information used for idea evaluation.
• output: Output results of the brainstorming process.
  • elapsed_time: Elapsed time for each step: (1) idea generation, (2) mindmap and class diagram generation, (3) idea evaluation in this order.
  • ideas: Generated ideas in dictionary format if multiple LLMs are used.
  • mindmap: Mindmap of the generated ideas in mermaid format.
  • class_diagram: Class diagram of the generated ideas in mermaid format.
  • evaluation: Evaluation of the generated ideas in dictionary format if multiple LLMs are used.

Status

On average, it takes about 30 seconds to complete a brainstorming process by monju.

There are many calls of print in the step-by-step example. Each print shows a different status, which is useful to know progress.

Types of bs.status are as follows:

• not_started: The brainstorming process has not started. Most of cases, this status is shown when the instance of Monju is created.
• idea_generation: Generating ideas in progress or finished.
• idea_organization: Organizing ideas in mindmap and class diagram in progress or finished.
• idea_evaluation: Evaluating ideas in progress or finished.
• verifying: Verifying if all the outputs are generated in progress or finished.
• done: The brainstorming process is completed.
• failed: The brainstorming process is failed at some step.

Supported LLM Providers

One of the most important features of Monju is to generate ideas from multiple LLMs.

Monju currently sets the following LLM providers as default of idea generation and evaluation:

• OpenAI: gpt-4o-mini
• Anthropic: claude-3-haiku-20240307
• Google: gemini-1.5-flash

And OpenAI gpt-4o is used for mindmap and class diagram generation.

Monju is capable of single or multiple LLMs for each step:

• Generate ideas (multiple/single LLM)
• Organize ideas in mindmap and class diagram (single LLM)
• Evaluate ideas (multiple/single LLM)

You can configure which providers and models to use for each step. Set arguments of LLM information in params dictionary when call generate_ideas(**params), organize_ideas(**params), and evaluate_ideas(**params). Not possible for verify() and brainstorming().

Note that providers are not limited to those shown above. Also possible to user other providers like MistralAI, Groq, and so on, as long as they are defined in LLMMaster class.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

Apache License 2.0