PHANTOM/README.md

PHANTOM: Pricing Heuristics Against Non-human Transaction Orchestration Mechanisms

Bachelor's Thesis Project by Daniel Rösel, IE University Madrid (2025) Advisor: Alberto Martín Izquierdo

Overview

PHANTOM is an academic research project investigating pricing heuristics to protect e-commerce platforms from exploitation by LLM agents in dynamic pricing environments. This project explores behavioral signature detection for agent identification and develops protection mechanisms against automated transaction orchestration.

Research Focus Areas:

• AI security in e-commerce systems
• Behavioral signature detection for autonomous agent identification
• Dynamic pricing protection mechanisms
• LLM agent behavior analysis and exploitation patterns

Project Structure

This repository contains both the research thesis and a full-stack experimental platform:

PHANTOM/
├── paper/              # LaTeX Bachelor's Thesis
│   ├── src/           # LaTeX source files
│   ├── build/         # Compiled PDF output
│   └── concat_code.sh # Auto-concatenate code for appendix
│
├── docs/              # GitHub Pages academic project page
│   └── index.html     # Academic project showcase
│
├── web/               # Next.js research platform dashboard
│   └── src/           # Frontend application
│
├── backend/           # Python backend services
│   ├── provider/      # Data provider service
│   └── worker/        # Kafka consumer worker
│
├── experiments/       # Research experiments and data analysis
│   └── data_export.ipynb
│
└── docker/            # Infrastructure configurations

Quick Start

Prerequisites

• Docker & Docker Compose (for infrastructure)
• Node.js 18+ (for web application)
• Python 3.9+ (for backend services)
• LaTeX distribution (for building thesis paper)

Infrastructure Setup

Start Kafka, Redis, and monitoring services:

docker-compose up -d

Services:

• Redis: localhost:6379
• Kafka: localhost:9092
• Zookeeper: localhost:2181
• Redpanda Console (Kafka UI): http://localhost:8080

Web Application

cd web
npm install
npm run dev

Access at http://localhost:3000

Backend Services

Install dependencies:

pip install -r requirements.txt

Run worker:

cd backend/worker
python main.py

Building the Thesis Paper

Using Make

make pdf        # Compile LaTeX to PDF
make watch      # Continuous compilation (live preview)
make clean      # Remove build artifacts

Manual Build

cd paper/src
latexmk -pdf main.tex

Output: paper/build/main.pdf

Automated CI/CD

The thesis PDF is automatically built via GitHub Actions on every push to main that affects paper/**. The compiled PDF artifact is available in the Actions tab.

Technical Architecture

Frontend: Next.js 14, React 18, TypeScript, Tailwind CSS Backend: Python, Kafka (event streaming) Infrastructure: Redis (cache), Kafka + Zookeeper Monitoring: Redpanda Console Data Analysis: Jupyter Notebooks, Pandas, Matplotlib

Event-Driven Architecture

The platform uses Kafka for real-time event streaming, enabling:

• Asynchronous task processing
• Scalable data collection
• Experiment tracking and analysis
• Behavioral pattern detection

Research Experiments

Jupyter notebooks in experiments/ contain:

• Data exploration and analysis
• Behavioral pattern visualizations
• Statistical analysis of agent behaviors
• Experiment result processing

Run experiments:

cd experiments
jupyter notebook data_export.ipynb

Documentation

• Academic Project Page: Hosted on GitHub Pages at /docs
• Thesis Paper: Latest PDF available via GitHub Actions artifacts
• Web App README: See web/README.md
• Backend READMEs: See backend/provider/README.md and backend/worker/README.md

Development Workflow

1. Paper Development: Edit LaTeX files in paper/src/, use make watch for live preview
2. Web Development: Standard Next.js workflow in web/
3. Backend Development: Python services in backend/
4. Experiments: Jupyter notebooks in experiments/

Code in Thesis Appendix

The paper/concat_code.sh script automatically generates a LaTeX appendix containing all source code from:

• backend/ (Python, JavaScript, Shell, YAML)
• experiments/ (Analysis scripts)
• docker/ (Infrastructure configs)
• web/src/ (TypeScript/React components)

This runs automatically during PDF compilation.

Contributing

This is an academic thesis project. For questions or collaboration inquiries, please open an issue.

License

Academic research project - all rights reserved.

Citation

@thesis{rosel2025phantom,
  title={Pricing Heuristics Against Non-human Transaction Orchestration Mechanisms},
  author={Rösel, Daniel},
  year={2025},
  school={IE University},
  address={Madrid, Spain},
  type={Bachelor's Thesis}
}

Acknowledgments

Special thanks to Alberto Martín Izquierdo for academic supervision and guidance throughout this research project.