Panel Of Claudes: Multi-Perspective AI Panel System

November 17, 2025 · View on GitHub

Use Cases: Multi-Perspective Analysis | Idea Exploration | Critical Thinking | Research Synthesis | Decision Support

Overview

Panel Of Claudes is an experimental multi-agent AI system that simulates expert panel discussions where multiple agents analyze topics through different analytical lenses. Rather than adversarial debate, the system creates a think tank environment where diverse perspectives illuminate different dimensions of complex topics.

Think of it as assembling a panel of domain experts—an economist, ethicist, scientist, and policy analyst—each bringing their unique framework to bear on the same question. The result isn't about "winning" an argument, but rather achieving a richer, multi-dimensional understanding.

Why This Matters

Multi-Lens Analysis: Topics explored through complementary frameworks (economic, ethical, scientific, etc.)
Beyond Binary Thinking: Escapes for/against limitations to reveal nuanced dimensions
Expert Panel Simulation: Mimics real-world research symposiums and policy roundtables
Productive Tensions: Identifies where perspectives conflict and where they align
Comprehensive Synthesis: A moderator weaves perspectives into actionable insights
Multi-Format Output: From machine-readable JSON to podcast-ready audio narratives

From Debate to Multi-Perspective Analysis

Traditional debate formats (pro vs. con) force complex topics into binary frames. Real-world decision-making rarely works this way. Instead, we need to understand:

How different stakeholders view the issue
What trade-offs exist between competing values
Where unexpected synergies might emerge
Which dimensions of a problem we hadn't considered

This system embraces that complexity by letting multiple perspectives coexist, interact, and inform each other.

System Architecture

flowchart TB
    User[User: Submit Complex Motion] --> Decompose[Motion Decomposition Agent]

    Decompose -->|Problem statements<br/>Causal claims<br/>Context| Auto[Auto-Perspective Selection]

    Auto -->|Suggested perspectives| Config[User Confirms/Modifies]

    Config --> R1[ROUND 1: Independent Analysis]

    R1 --> P1[Economic Lens]
    R1 --> P2[Political Lens]
    R1 --> P3[Social Lens]
    R1 --> P4[Educational Lens]
    R1 --> P5[Perspective N]

    P1 -->|Deep analysis| Interim[Interim Synthesis Agent]
    P2 -->|Deep analysis| Interim
    P3 -->|Deep analysis| Interim
    P4 -->|Deep analysis| Interim
    P5 -->|Deep analysis| Interim

    Interim -->|Tensions<br/>Gaps<br/>Questions| R2[ROUND 2: Cross-Pollination]

    R2 --> P1B[Economic Lens<br/>Responds & Refines]
    R2 --> P2B[Political Lens<br/>Responds & Refines]
    R2 --> P3B[Social Lens<br/>Responds & Refines]
    R2 --> P4B[Educational Lens<br/>Responds & Refines]
    R2 --> P5B[Perspective N<br/>Responds & Refines]

    P1B --> Moderator[Moderator: Final Synthesis]
    P2B --> Moderator
    P3B --> Moderator
    P4B --> Moderator
    P5B --> Moderator

    Moderator --> Report[Comprehensive Report]

    Report --> JSON[JSON Output]
    Report --> MD[Markdown Report]

    MD --> PDF[PDF Document]
    MD --> EPUB[ePub Document]

    Report --> Formatter[SSML Formatter Agent]
    Formatter --> TTS[Text-to-Speech]
    TTS --> Podcast[Podcast Audio]

    style User fill:#e1f5ff
    style Decompose fill:#ffeaa7
    style R1 fill:#d4edda
    style Interim fill:#fff3cd
    style R2 fill:#d1ecf1
    style Moderator fill:#ffeaa7
    style Report fill:#dfe6e9
    style Podcast fill:#e7e7ff

Perspective Lens Library

graph TB
    subgraph "Core Perspectives"
        P1[Economic Lens]
        P2[Ethical Lens]
        P3[Scientific Lens]
        P4[Environmental Lens]
        P5[Social Lens]
        P6[Political Lens]
    end

    subgraph "Academic Perspectives"
        A1[Historical Lens]
        A2[Philosophical Lens]
        A3[Sociological Lens]
        A4[Psychological Lens]
    end

    subgraph "Business Perspectives"
        B1[Financial Lens]
        B2[Strategic Lens]
        B3[Operational Lens]
        B4[Customer-Centric Lens]
    end

    subgraph "Policy Perspectives"
        L1[Legal/Regulatory Lens]
        L2[Implementation Lens]
        L3[Social Impact Lens]
        L4[Risk Management Lens]
    end

    style P1 fill:#d4edda
    style P2 fill:#fff3cd
    style P3 fill:#d1ecf1
    style P4 fill:#f8d7da
    style P5 fill:#e8daef
    style P6 fill:#ffeaa7

How It Works

Panel Configuration Modes

1. Fixed Perspectives Mode

User selects 3-7 specific lenses from the perspective library
Best for targeted analysis of known dimensions

2. Auto-Perspective Mode

System analyzes the topic and selects relevant perspectives
Best for exploratory analysis

3. Custom Perspectives Mode

User defines their own analytical frameworks
Best for specialized domains or unique analytical needs

Sequential Flow

sequenceDiagram
    participant User
    participant Decompose as Decomposition Agent
    participant Auto as Auto-Perspective
    participant P1 as Economic Lens
    participant P2 as Political Lens
    participant P3 as Social Lens
    participant Interim as Interim Synthesis
    participant Moderator
    participant Output

    User->>Decompose: Submit complex motion
    Note over Decompose: Extract problems,<br/>causal claims,<br/>implicit questions

    Decompose->>Auto: Structured components
    Note over Auto: Analyze & suggest<br/>relevant perspectives

    Auto->>User: Recommended perspectives
    User->>P1: Confirm & distribute
    User->>P2: Confirm & distribute
    User->>P3: Confirm & distribute

    Note over P1: ROUND 1:<br/>Deep independent<br/>analysis

    Note over P2: ROUND 1:<br/>Deep independent<br/>analysis

    Note over P3: ROUND 1:<br/>Deep independent<br/>analysis

    P1->>Interim: Round 1 analysis
    P2->>Interim: Round 1 analysis
    P3->>Interim: Round 1 analysis

    Note over Interim: Identify tensions,<br/>gaps, generate<br/>cross-perspective questions

    Interim->>P1: Questions for Economic lens
    Interim->>P2: Questions for Political lens
    Interim->>P3: Questions for Social lens

    Note over P1: ROUND 2:<br/>Respond to others,<br/>refine analysis

    Note over P2: ROUND 2:<br/>Respond to others,<br/>refine analysis

    Note over P3: ROUND 2:<br/>Respond to others,<br/>refine analysis

    P1->>Moderator: Refined analysis
    P2->>Moderator: Refined analysis
    P3->>Moderator: Refined analysis

    Note over Moderator: Final synthesis:<br/>Validate problems,<br/>map causation,<br/>identify solutions

    Moderator->>Output: Generate reports<br/>(JSON, MD, PDF, ePub)
    Moderator->>Output: Create podcast version<br/>(SSML → TTS)

Phase 0: Motion Decomposition

Input: Complex motion or problem statement from user

Process: Decomposition agent analyzes the motion and extracts:

problem_statements:
  - Core issues identified in the motion
  - Quantifiable concerns
  - Observed patterns

causal_claims:
  - "X leads to Y" statements
  - Proposed mechanisms
  - Feedback loops identified

implicit_questions:
  - What questions is this motion really asking?
  - What decisions need to be made?
  - What trade-offs exist?

context:
  - Geographic/cultural setting
  - Relevant domains
  - Stakeholder groups
  - Historical background

Output: Structured motion components that become inputs for perspective agents

Phase 1: Auto-Perspective Recommendation

Input: Decomposed motion components

Process: System analyzes motion structure and suggests:

required_perspectives:
  - Perspectives essential for this specific motion
  - Based on domains identified in context

recommended_perspectives:
  - Perspectives that would add significant value
  - Optional but suggested

optional_perspectives:
  - Additional lenses that could provide insight
  - User can add if interested

User Action: Confirms, modifies, or manually selects perspectives

Output: Finalized perspective configuration (typically 3-7 lenses)

Phase 2: Round 1 - Independent Deep Analysis

Each perspective agent receives structured instructions:

Mission: Comprehensive understanding through your analytical lens

Input Provided to Each Agent:

Decomposed motion (problems, causal claims, questions, context)
Their specific analytical framework
Structured analysis template

Analysis Framework for Each Perspective:

Problem Validation
- Which problem statements are valid through your lens?
- What evidence supports/contradicts them?
- What problems are missing?
Causal Analysis
- Evaluate each causal claim
- Identify alternative explanations
- Map feedback loops and systemic effects
Trade-offs & Tensions
- What values/priorities conflict?
- What are opportunity costs?
- What second-order effects matter?
Solution Spaces
- What interventions does your lens suggest?
- Implementation challenges?
- Success metrics?
Lens Limitations
- What can your lens NOT address?
- Where do you need other perspectives?
- What uncertainties remain?

Output: Each perspective produces comprehensive independent analysis

Phase 3: Interim Synthesis - Identifying Productive Tensions

Input: All Round 1 analyses

Process: Interim synthesis agent identifies:

convergent_themes:
  - Where perspectives agree
  - Confirmed problems across lenses
  - Aligned causal mechanisms

productive_tensions:
  - Where perspectives constructively disagree
  - Competing frameworks
  - Different value priorities
  - Questions that emerge from tensions

knowledge_gaps:
  - What hasn't been addressed
  - Empirical questions raised
  - Missing perspectives

cross_perspective_questions:
  - Specific questions for each lens based on others' analyses
  - Clarifications needed
  - Opportunities for refinement

Output:

Summary of convergences and tensions
Specific questions for each perspective to address in Round 2

Input to Each Perspective:

Their own Round 1 analysis
Summary of other perspectives' analyses
Tensions involving their perspective
Specific questions directed at them

Instructions:

Respond to questions directed at your lens
Acknowledge valid points from other perspectives
Clarify where others misunderstand your framework
Refine your analysis based on cross-perspective insights
Identify where perspectives complement each other

Output: Each perspective produces refined analysis that:

Responds to other perspectives
Refines original analysis
Acknowledges complementary insights
Maintains analytical framework integrity

Phase 5: Final Moderation & Synthesis

Input:

Round 1 independent analyses
Interim synthesis
Round 2 refined analyses

Process: Moderator creates comprehensive synthesis structured as:

Executive Summary
- Multi-paragraph overview of key findings
Problem Validation Across Perspectives
- Strongly confirmed problems
- Nuanced/contested issues
- Newly identified problems
Causal Analysis
- Competing explanations mapped
- Feedback loops identified
- Multi-perspective causal models
Productive Tensions
- Value conflicts between perspectives
- Trade-offs identified
- Resolution spaces explored
Solution Spaces
- Where perspectives converge on solutions
- Complementary approaches
- Implementation considerations
Knowledge Gaps
- Empirical questions remaining
- Normative questions identified
- Areas needing further analysis
Implications for Action
- Scenario-based recommendations
- Integrated approaches
- Decision frameworks

Output: Comprehensive multi-perspective report

Output Formats

Machine-Readable

JSON: Structured data for integration with other systems
- Topic metadata
- Selected perspectives and their frameworks
- Individual perspective analyses
- Cross-perspective insights and tensions
- Timestamps and flow

Human-Readable Documents

Markdown: Clean, formatted panel proceedings with perspective-by-perspective analysis
PDF: Print-ready multi-perspective report
ePub: E-reader friendly format for in-depth reading

Audio Content

SSML Formatting: Panel discussion formatted with podcast host personality
TTS Audio: Text-to-speech narration of the full multi-perspective analysis
Podcast: Ready-to-listen "think tank" style audio content

Implementation Options

Quick Start (Claude Code)

Use Claude Code with custom agents to implement a miniature version:

Define agent prompts
Chain agent interactions
Generate outputs

Full Implementation (Multi-Agent Frameworks)

Leverage frameworks like CrewAI, AutoGen, or LangGraph:

Code-defined agent behaviors
Persistent state management
Advanced orchestration
Custom output pipelines

Example: Israeli Socio-Economic Analysis

Motion Submitted:

"The State of Israel suffers from ongoing cost of living problems and severe income inequality. The transition from socialist economy to high-tech powerhouse created a binary economy where high-tech workers thrive while others struggle with wages that don't match living costs. The education system focuses on STEM to feed this single sector. Meanwhile, the political system prioritizes security over domestic concerns and lacks geographic accountability, as the concept of constituency representation is foreign to Israel."

Phase 0: Motion Decomposition

Problem Statements Extracted:

High cost of living relative to wages
Severe income inequality (among world's highest)
Economic over-dependence on high-tech sector
Binary/tiered economic structure
Political system lacks geographic accountability
Security concerns eclipse domestic policy priorities
Education system over-indexed on STEM

Causal Claims Identified:

High-tech economic focus → income inequality
Single-sector dominance → binary economy
STEM education focus → workforce pipeline for one sector
Political structure without geographic mandate → accountability deficit
Security prioritization → neglect of cost-of-living issues

Implicit Questions:

Is economic diversification desirable and achievable?
Would geographic constituency system improve accountability?
How should education balance economic utility vs. broader development?
Can domestic priorities coexist with security needs?

Phase 1: Auto-Perspective Recommendation

Required Perspectives:

Economic Lens (inequality, high-tech dominance)
Political Lens (governance structure, accountability)
Social Lens (cost of living, class divisions)

Recommended Perspectives:

Educational Lens (STEM focus, workforce development)
Geopolitical Lens (security vs. domestic priorities)

Optional Perspectives:

Historical Lens (socialist → capitalist transition)
Philosophical Lens (meritocracy vs. equity values)

User Selects: Economic, Political, Social, Educational, Geopolitical

Phase 2: Round 1 - Sample Independent Analysis

Economic Lens (excerpts):

✅ Confirms income inequality is severe (Gini coefficient data)
✅ Confirms over-indexing on single sector (concentration risk)
⚠️ Challenges "binary" framing—sees multi-tiered economy
Causal analysis: High-tech creates wage ceiling; floor set by policy, housing, education access
Solution space: Economic diversification incentives, progressive taxation, housing supply interventions

Political Lens (excerpts):

✅ Confirms accountability deficit in current system
Causal analysis: List-based proportional representation → parties accountable to party leadership, not citizens
Geographic constituencies: Could create local accountability for housing, cost-of-living issues
⚠️ Trade-off: Might weaken national consensus on security policy

Social Lens (excerpts):

✅ Confirms binary experience ("high-tech vs. everyone else")
New problem identified: Social cohesion degradation, class resentment
Causal analysis: Economic inequality → residential segregation → reduced social mixing
Limitation: Cannot evaluate technical economic mechanisms

Phase 3: Interim Synthesis

Convergent Themes:

All perspectives confirm inequality problem
Economic, Social, Educational lenses identify self-reinforcing cycles
Multiple perspectives see housing as critical variable

Productive Tensions:

Economic vs. Political: Is inequality economic inevitability or policy failure?
Educational vs. Economic: Should education serve economy or human development?
Geopolitical vs. Social: How to balance security spending with domestic investment?

Questions for Round 2:

To Economic: If political structure changed, would economic outcomes change?
To Political: Do other countries with geographic constituencies show better outcomes?
To Geopolitical: Does inequality itself pose security risks?

Phase 4: Round 2 - Sample Refined Analysis

Economic Lens Response:

Acknowledges Political lens point: Market outcomes aren't natural laws—they respond to institutional frameworks
Refinement: Round 1 understated role of policy choices
Responds to Geopolitical: Can model security-inequality feedback loop

Political Lens Response:

Responds to empirical question: Points to constituency systems in comparable democracies
Acknowledges Economic lens: Geographic reform alone won't solve inequality—needs complementary policies
Clarifies to Geopolitical: Local accountability doesn't preclude national security consensus

Phase 5: Final Synthesis (Excerpt)

Problem Validation:

Strongly Confirmed: Income inequality (all lenses), cost of living crisis (Economic, Social), accountability deficit (Political, Social)
Nuanced: "Binary economy" framing—Economic sees multi-tier structure; Social sees binary lived experience

Causal Analysis - High-Tech Dominance → Inequality:

Economic: Market-rational outcome given incentive structures
Political: Policy choice to favor sector through tax breaks, R&D investment
Educational: Education system responds to labor market signals
Synthesis: Self-reinforcing cycle where all three factors interact via feedback loops

Productive Tension - Economic Efficiency vs. Social Equity:

Economic perspective values growth, innovation, global competitiveness
Social perspective values equality, quality of life, social cohesion
Tension point: High-tech success drives GDP growth BUT increases inequality
Resolution space: Progressive redistribution + social investment funded by growth

Solution Convergence - Economic Diversification:

Economic: Tax incentives for other sectors, R&D support
Political: Regional development tied to constituency interests
Educational: Broader curricula creating versatile workforce
Integrated approach: All mechanisms could work together

Knowledge Gaps:

Empirical: Have geographic constituencies in similar countries improved outcomes?
Normative: What is proper role of education—economic utility or human development?
Practical: What is citizen appetite for economic vs. security prioritization?

Future Enhancements

Perspective Persona Specialization

Perspectives adopt specific schools of thought (e.g., "Keynesian Economic Lens" vs "Austrian Economic Lens")
Historical figure perspectives (e.g., "What would Carl Sagan's perspective be?")
Domain expert simulation for specialized topics

Multi-Panel Conferences

Chain multiple panels together with different perspective sets
Output from one panel feeds into another (e.g., high-level analysis → implementation perspectives)
Simulate multi-day academic conference structures
Progressive refinement through iterative panels

Interactive Perspective Selection

AI-assisted perspective recommendation based on topic analysis
User can add/remove perspectives mid-analysis
Dynamic perspective weighting based on topic priorities
Community-contributed custom perspective frameworks

Live Interaction

Stream perspective analyses in real-time
Avatar visualization for each perspective agent
Interactive audience participation and questions
Live moderator synthesis as perspectives develop

Use Cases

Strategic Decision Making: Evaluate complex decisions through multiple analytical frameworks
Policy Analysis: Understand multi-dimensional implications of policy proposals
Research Synthesis: Explore academic topics from complementary disciplinary perspectives
Content Creation: Generate nuanced, multi-perspective content for publications
Education: Teach multi-dimensional thinking and analytical framework application
Due Diligence: Comprehensive analysis of business opportunities, investments, or initiatives
Ethics Committees: Simulate multi-stakeholder perspectives for ethical deliberation
Innovation Workshops: Explore ideas through diverse lenses to identify opportunities and risks

Getting Started

Clone this repository
Review the perspective lens library in /perspectives
Configure your preferred multi-agent framework (CrewAI, AutoGen, LangGraph)

Select your perspectives for analysis:

python main.py --topic "Your topic here" --perspectives economic,ethical,scientific

Or use auto-mode:

python main.py --topic "Your topic here" --auto-perspectives

Review multi-perspective outputs in /output

Technical Stack

Language: Python
Framework: CrewAI / AutoGen (configurable)
Output Generation:
- Markdown: Python-Markdown
- PDF: WeasyPrint / ReportLab
- ePub: ebooklib
- TTS: Google TTS / Azure Speech / ElevenLabs

Contributing

Contributions welcome! Areas of interest:

Additional output formats
Enhanced agent personalities
Integration with research databases
UI/UX improvements

License

MIT License - See LICENSE file for details

Author

Daniel Rosehill Technology Communications Specialist | Automation Expert danielrosehill.com

To view an index of my Claude Code related projects, click here

"We don't see things as they are, we see them as we are." - Anaïs Nin

Panel Of Claudes helps us see things as they could be—through many lenses at once.