🤖 AI Workforce of the Future

Comparative Analysis of GraphRAG-Enhanced Multi-Agent Systems for Qualitative Organizational Research

📋 Table of Contents

Overview
Research Context
Features
Autonomy Stages
Agent Pool
System Architecture
Technology Stack
Getting Started
Configuration
API Reference
Contributing
License
Academic Attribution

🎯 Overview

This research platform is a comprehensive tool for comparing AI autonomy levels in organizational analysis contexts. It implements a progression from simple reactive chatbots to sophisticated GraphRAG-enhanced multi-agent systems, enabling rigorous comparative analysis of AI capabilities for qualitative research tasks.

🎓 Academic Project
Developed as part of a Master's thesis at Aalto University School of Business, focusing on how different AI autonomy levels and multi-agent architectures can enhance qualitative organizational research.

📚 Research Context

Thesis Information

Field	Details
Title	Enhancing Qualitative Organizational Analysis through Integrated GraphRAG and Multi-Agent AI Systems
Author	Ali Amaan
Programme	Master's Programme in Economics & Business Administration
Major	International Design Business Management (IDBM)
Institution	Aalto University School of Business
Partner	Department of Management Studies, Aalto University
Expected Completion	December 2025

Research Questions

How do different AI autonomy levels compare in supporting qualitative organizational analysis?
What role does GraphRAG play in enhancing multi-agent system memory and context retention?
How can human-in-the-loop (HIL) mechanisms improve AI-assisted research workflows?

Keywords

GraphRAG Multi-Agent Systems Knowledge Graphs Qualitative Analysis Organizational Research Agentic AI Human-in-the-Loop LLM Orchestration

✨ Features

🧠 Intelligent Analysis

5 Autonomy Levels — From reactive chatbot to autonomous multi-agent systems
12 Specialized Agents — Manager, Researcher, Analyst, Visualizer, Engineer, and more
Code Interpreter — Execute Python for data analysis and visualization
Web Search — Real-time information retrieval via DuckDuckGo
Image Generation — AI-powered visual content creation

📊 GraphRAG Integration

Knowledge Graph Memory — Persistent entity and relationship storage
Semantic Search — Find relevant facts and node summaries
Episode Management — Add, retrieve, and manage conversation episodes
Entity Extraction — Automatic extraction of entities from conversations
Temporal Awareness — Track when information was learned

🎨 Modern UI/UX

Real-time Streaming — WebSocket-based response streaming
Neo4j Visualization — Interactive knowledge graph explorer
Plotly Charts — Rich data visualizations
Dark Theme — Beautiful, researcher-friendly interface
Responsive Design — Works on desktop and tablet

🔧 Developer Experience

Type-Safe — Full TypeScript frontend, Pydantic backend
Hot Reload — Vite for frontend, Uvicorn for backend
Modular Architecture — Clean separation of concerns
Trace Logging — Detailed agent execution traces
Session Management — Persistent analysis sessions

🎚️ Autonomy Stages

The platform implements a Sheridan Scale-inspired progression of AI autonomy levels:

graph LR
    subgraph "Problem Space"
        S1[🗨️ Stage 1<br/>Reactive Baseline]
        S2[🔧 Stage 2<br/>Tool-Augmented]
    end
    
    subgraph "Collaboration Space"
        S3[👥 Stage 3<br/>HIL Single-Agent]
    end
    
    subgraph "Autonomy Space"
        S4[🤖 Stage 4<br/>Autonomous MAS]
        S5[🌐 Stage 5<br/>HIL Multi-Agent]
    end
    
    S1 --> S2 --> S3 --> S4 --> S5
    
    style S1 fill:#3B82F6,stroke:#1E40AF,color:#fff
    style S2 fill:#F59E0B,stroke:#B45309,color:#fff
    style S3 fill:#F97316,stroke:#C2410C,color:#fff
    style S4 fill:#EF4444,stroke:#B91C1C,color:#fff
    style S5 fill:#8B5CF6,stroke:#6D28D9,color:#fff

Stage	Name	Description	Sheridan Level	Capabilities
1	Reactive Baseline	Stateless LLM with minimal context	Level 1	Basic Q&A, no memory
2	Tool-Augmented Agent	Single agent with tools	Level 5-6	Code Interpreter, Web Search
3	HIL Single-Agent	Human-supervised single agent	Level 5-6	Human approval, transparent reasoning
4	Autonomous MAS	Self-directed multi-agent system	Level 8-10	Parallel agent execution, quality loops
5	HIL Multi-Agent	Human-in-the-loop multi-agent	Level 5-7	Team collaboration, human oversight

🤝 Agent Pool

The multi-agent stages (4 & 5) leverage a pool of 12 specialized agents that work together using an agents-as-tools pattern:

graph TB
    subgraph Manager["👔 Manager (Orchestrator)"]
        M[Project Manager]
    end
    
    subgraph Specialists["Specialist Agents"]
        R[🔍 Researcher<br/>Web Search]
        A[📊 Data Analyst<br/>Code Interpreter]
        V[📈 Visualizer<br/>Charts & Plots]
        G[🎨 Graphic Artist<br/>Image Generation]
        E[💻 Engineer<br/>Code Interpreter]
        W[✍️ Writer<br/>Documentation]
        C[🔎 Critic<br/>Quality Review]
        S[🎯 Strategist<br/>Business Analysis]
        D[🎓 Domain Expert<br/>Subject Matter]
        AS[🤝 Assistant<br/>General Support]
    end
    
    M --> R
    M --> A
    M --> V
    M --> G
    M --> E
    M --> W
    M --> C
    M --> S
    M --> D
    M --> AS
    
    style M fill:#3B82F6,stroke:#1E40AF,color:#fff
    style R fill:#10B981,stroke:#047857,color:#fff
    style A fill:#F59E0B,stroke:#B45309,color:#fff
    style V fill:#8B5CF6,stroke:#6D28D9,color:#fff
    style G fill:#E91E63,stroke:#AD1457,color:#fff
    style E fill:#EC4899,stroke:#BE185D,color:#fff
    style W fill:#06B6D4,stroke:#0E7490,color:#fff
    style C fill:#EF4444,stroke:#B91C1C,color:#fff
    style S fill:#F97316,stroke:#C2410C,color:#fff
    style D fill:#84CC16,stroke:#4D7C0F,color:#fff
    style AS fill:#64748B,stroke:#475569,color:#fff

Agent Capabilities

Agent	Role	Tools	Specialization
👔 Manager	Orchestrator	—	Task decomposition, delegation, synthesis
🔍 Researcher	Research Analyst	Web Search	Information gathering, source evaluation
📊 Data Analyst	Quantitative Analyst	Code Interpreter	Statistical analysis, pattern detection
📈 Visualizer	Visualization Specialist	Code Interpreter, Image Gen	Charts, graphs, dashboards
🎨 Graphic Artist	Visual Concept Artist	Image Generation	Illustrations, concept art
💻 Engineer	Software Engineer	Code Interpreter	Algorithms, data pipelines
✍️ Writer	Technical Writer	—	Reports, documentation
🔎 Critic	Quality Reviewer	—	Error detection, improvements
🎯 Strategist	Business Strategist	Web Search	Strategic insights, recommendations
🎓 Domain Expert	Subject Matter Expert	Web Search, File Search	Domain knowledge, validation
🤝 Assistant	General Assistant	Web Search, Code Interpreter	Flexible support tasks

🏗️ System Architecture

graph TB
    subgraph Frontend["🖥️ Frontend (React + TypeScript)"]
        UI[UI Components]
        WS[WebSocket Client]
        Store[Zustand Store]
        GraphViz[Neo4j Visualizer]
    end
    
    subgraph Backend["⚙️ Backend (FastAPI + Python)"]
        API[REST API]
        WSS[WebSocket Server]
        Stages[Stage Implementations]
        Tools[Tool Layer]
    end
    
    subgraph GraphitiMCP["🧠 Graphiti MCP Server"]
        MCP[MCP Protocol Handler]
        GraphCore[Graphiti Core]
        Episodes[Episode Management]
        Search[Semantic Search]
    end
    
    subgraph Database["💾 Neo4j Database"]
        Nodes[(Entity Nodes)]
        Edges[(Relationships)]
        Embeddings[(Vector Embeddings)]
    end
    
    subgraph External["☁️ External Services"]
        OpenAI[OpenAI API<br/>GPT-5]
        DDG[DuckDuckGo<br/>Web Search]
    end
    
    UI --> Store
    Store --> WS
    WS <--> WSS
    UI --> API
    API --> Stages
    Stages --> Tools
    Tools --> OpenAI
    Tools --> DDG
    Stages --> MCP
    MCP --> GraphCore
    GraphCore --> Nodes
    GraphCore --> Edges
    GraphCore --> Embeddings
    GraphViz --> Nodes
    
    style Frontend fill:#1E293B,stroke:#475569,color:#fff
    style Backend fill:#1E293B,stroke:#475569,color:#fff
    style GraphitiMCP fill:#1E293B,stroke:#475569,color:#fff
    style Database fill:#1E293B,stroke:#475569,color:#fff
    style External fill:#1E293B,stroke:#475569,color:#fff

🛠️ Technology Stack

Frontend

Technology	Version	Purpose
React	19.2.0	UI framework
TypeScript	5.9.3	Type safety
Vite	7.2.4	Build tool & dev server
TailwindCSS	3.4.14	Styling
Framer Motion	12.23.26	Animations
Zustand	5.0.9	State management
React Router	7.11.0	Navigation
@neo4j-nvl/react	1.0.0	Graph visualization
Plotly.js	3.3.1	Data visualization

Backend

Technology	Version	Purpose
Python	3.10+	Runtime
FastAPI	0.115.5	API framework
Uvicorn	0.32.0	ASGI server
OpenAI SDK	2.9.0+	LLM integration
OpenAI Agents SDK	0.0.16+	Multi-agent orchestration
Pandas	2.2.3	Data processing
Pydantic	2.12.3+	Data validation
WebSockets	12.0	Real-time communication

Knowledge Graph

Technology	Version	Purpose
Graphiti Core	Latest	GraphRAG framework
Graphiti MCP Server	Latest	Model Context Protocol
Neo4j	5.26+	Graph database

🚀 Getting Started

Prerequisites

Before you begin, ensure you have the following installed:

Node.js 20+ and npm (or yarn/pnpm)
Python 3.10+ with pip or uv
Docker and Docker Compose (for Neo4j)
OpenAI API Key with GPT-5 access

⚠️ Important: This platform is designed to work with GPT-5 for optimal performance. Lower-tier models may not support all features (reasoning effort, extended thinking, etc.).

Backend Setup

Navigate to the backend directory:

cd backend

Create and activate a virtual environment:

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

Install dependencies:

pip install -r requirements.txt

Configure environment variables:

cp env.example .env

Edit .env with your configuration:

# Required
OPENAI_API_KEY=sk-your-api-key-here
OPENAI_MODEL=gpt-5

# Optional GPT-5 tuning
OPENAI_REASONING_EFFORT=high
OPENAI_VERBOSITY=high

# Session & Data
SESSION_DIR=./backend/sessions
DATASET_DIR=./backend/datasets
LOG_LEVEL=INFO
ENABLE_AGENT_TRACES=1

# Graphiti Knowledge Graph
GRAPHITI_MCP_URL=http://localhost:8000/mcp/
GRAPHITI_ENABLED=1

# Neo4j Database
NEO4J_URI=bolt://localhost:7687
NEO4J_USER=neo4j
NEO4J_PASSWORD=your_secure_password

Start the backend server:

uvicorn app.main:app --host 0.0.0.0 --port 8080 --reload

The API will be available at http://localhost:8080

Graphiti MCP & Neo4j Setup

Navigate to the Graphiti MCP directory:

cd graphiti/mcp_server

Start Neo4j and MCP Server using Docker Compose:

# Using Neo4j (recommended for this project)
docker compose -f docker/docker-compose-neo4j.yml up -d

Or if you prefer FalkorDB:

docker compose up -d

Verify the services are running:

# Check MCP Server health
curl http://localhost:8000/health

# Access Neo4j Browser
open http://localhost:7474

Configure your .env file in mcp_server/:

OPENAI_API_KEY=sk-your-api-key-here
NEO4J_URI=bolt://localhost:7687
NEO4J_USER=neo4j
NEO4J_PASSWORD=your_secure_password

Frontend Setup

Navigate to the frontend directory:

cd frontend

Install dependencies:

npm install

Start the development server:

npm run dev

The application will be available at http://localhost:5173

Quick Start (All Services)

For convenience, you can start all services together:

# Terminal 1: Neo4j + Graphiti MCP
cd graphiti/mcp_server && docker compose -f docker/docker-compose-neo4j.yml up

# Terminal 2: Backend API
cd backend && uvicorn app.main:app --host 0.0.0.0 --port 8080 --reload

# Terminal 3: Frontend
cd frontend && npm run dev

⚙️ Configuration

Environment Variables Reference

Variable	Required	Default	Description
`OPENAI_API_KEY`	✅	—	Your OpenAI API key
`OPENAI_MODEL`	❌	`gpt-5`	Model to use for LLM calls
`OPENAI_REASONING_EFFORT`	❌	`high`	Reasoning depth (low/medium/high)
`OPENAI_VERBOSITY`	❌	`high`	Response verbosity
`SESSION_DIR`	❌	`./backend/sessions`	Session storage path
`DATASET_DIR`	❌	`./backend/datasets`	Dataset storage path
`LOG_LEVEL`	❌	`INFO`	Logging level
`ENABLE_AGENT_TRACES`	❌	`1`	Enable detailed agent traces
`GRAPHITI_MCP_URL`	❌	`http://localhost:8000/mcp/`	Graphiti MCP endpoint
`GRAPHITI_ENABLED`	❌	`1`	Enable/disable GraphRAG
`NEO4J_URI`	❌	`bolt://localhost:7687`	Neo4j Bolt URI
`NEO4J_USER`	❌	`neo4j`	Neo4j username
`NEO4J_PASSWORD`	❌	—	Neo4j password

Graphiti MCP Configuration

The Graphiti MCP server can be configured via config.yaml:

server:
  transport: "http"  # Options: http, stdio

llm:
  provider: "openai"  # Options: openai, anthropic, gemini, groq
  model: "gpt-5"

database:
  provider: "neo4j"  # Options: neo4j, falkordb
  providers:
    neo4j:
      uri: "${NEO4J_URI:bolt://localhost:7687}"
      username: "${NEO4J_USER:neo4j}"
      password: "${NEO4J_PASSWORD}"

graphiti:
  entity_types:
    - name: "Preference"
      description: "User preferences, choices, opinions"
    - name: "Organization"
      description: "Companies, institutions, groups"
    - name: "Topic"
      description: "Subjects of conversation or interest"

📡 API Reference

REST Endpoints

Method	Endpoint	Description
`GET`	`/health`	Health check
`POST`	`/session/new`	Create new session
`DELETE`	`/session/{id}`	Delete session
`POST`	`/session/{id}/reset`	Reset session
`GET`	`/session/{id}/suggested_tasks`	Get AI-suggested tasks
`POST`	`/datasets`	Upload dataset
`GET`	`/datasets`	List all datasets
`GET`	`/api/graph/status`	Check knowledge graph status
`GET`	`/api/graph/data`	Get graph visualization data
`GET`	`/api/agents/pool`	Get available agents

WebSocket Protocol

Connect to /ws/{session_id} for real-time communication:

// Message types
type WSMessage = {
  type: "start" | "continue" | "cancel";
  task?: string;
  stage?: number;
  reasoning_effort?: string;
};

// Response events
type WSEvent = {
  type: "text" | "thinking" | "tool_call" | "file" | "error" | "done";
  content?: string;
  metadata?: object;
};

🤝 Contributing

We welcome contributions that align with the research goals of this project. Please see CONTRIBUTING.md for detailed guidelines.

⚠️ Note: This is an academic research project. All contributions must comply with the licensing terms and academic integrity requirements.

📄 License

This project is proprietary software developed as part of academic research at Aalto University. See LICENSE for full terms.

Restrictions:

❌ No redistribution without explicit permission
❌ No commercial use
❌ No derivative works without attribution
✅ Academic citation permitted with proper attribution

🎓 Academic Attribution

If you reference this work in academic publications, please use the following citation:

@mastersthesis{amaan2025graphrag,
  author       = {Ali Amaan},
  title        = {Enhancing Qualitative Organizational Analysis through Integrated GraphRAG and Multi-Agent AI Systems},
  school       = {Aalto University School of Business},
  year         = {2025},
  type         = {Master's Thesis},
  address      = {Espoo, Finland},
  month        = {December},
  note         = {International Design Business Management (IDBM) Programme}
}

_{Built with ❤️ at Aalto University School of Business}

Name		Name	Last commit message	Last commit date
Latest commit History 1 Commit
backend		backend
frontend		frontend
graphiti		graphiti
public		public
sample_data		sample_data
.gitignore		.gitignore
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
README.md		README.md

Folders and files

Latest commit

History

Repository files navigation

🤖 AI Workforce of the Future

📋 Table of Contents

🎯 Overview

📚 Research Context

Thesis Information

Research Questions

Keywords

✨ Features

🧠 Intelligent Analysis

📊 GraphRAG Integration

🎨 Modern UI/UX

🔧 Developer Experience

🎚️ Autonomy Stages

🤝 Agent Pool

Agent Capabilities

🏗️ System Architecture

🛠️ Technology Stack

Frontend

Backend

Knowledge Graph

🚀 Getting Started

Prerequisites

Backend Setup

Graphiti MCP & Neo4j Setup

Frontend Setup

Quick Start (All Services)

⚙️ Configuration

Environment Variables Reference

Graphiti MCP Configuration

📡 API Reference

REST Endpoints

WebSocket Protocol

🤝 Contributing

📄 License

🎓 Academic Attribution

About

Topics

Resources

License

Contributing

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages