About

About this site

The companion site of the master's thesis From Reactive Control to Holistic Planning: Evolutionary Prompt Optimization for Hybrid Deterministic-Neural Agents in Long-Horizon Environments (Master's thesis, M.Sc. Data Science, Leuphana University Lüneburg), supervised by Prof. Dr. Ricardo Usbeck and Prof. Dr. Burkhardt Funk.

What is here

Every result file and stored episode behind the thesis tables, replayed in the browser; the exact solver and the engine ported to TypeScript; the prompt-optimisation runs; the full text of the thesis. Nothing on this site calls a language model: it is a static replay of the stored evidence.

The framework that produced the results, with the evaluation platform that runs new experiments, is mt-ai-lab. The site's own code is mt-thesis.

Thesis PDFRead online

Imprint

Responsible for the content: King Wai Siu, student of M.Sc. Data Science at Leuphana University Lüneburg.

Contact: king.w.siu@stud.leuphana.de

This is a private, non-commercial academic project. It is not an official publication of the university.

Privacy notice

Hosting. The site is a set of static files served by Vercel Inc. To deliver them, Vercel processes the technical data every web server receives (IP address, time, requested page, browser) and keeps it in short-lived logs for operation and security.

Visitor statistics. Vercel Web Analytics counts page views without cookies. Each view is recorded anonymously, not tied to a person or an IP address, with its time, page, referrer, approximate location, operating system, browser and device type; visitors are told apart only by a hash of the request that is discarded after 24 hours.

In your browser. The theme, the state of the side rail and the game's saves are stored in your browser's local storage and never leave your device.

Legal basis: legitimate interest in running and improving the site (Art. 6(1)(f) GDPR). You have the rights of access, rectification, erasure, restriction and objection, and may complain to a data-protection supervisory authority. Contact as in the imprint.

Licences

  • Code of this site: MIT.
  • Result files, episode logs and text written for the thesis: CC BY 4.0.
  • The 24-floor game, its tiles and data: BSD 3-Clause, ckcz123/mota-js.
  • The single-floor test maps: from ckcz123/magic-tower-AI, which states no licence; shown with attribution.

Details in THIRD_PARTY.md.

Cite

Siu, K. W. (2026). From Reactive Control to Holistic Planning: Evolutionary Prompt Optimization for Hybrid Deterministic-Neural Agents in Long-Horizon Environments. Master's thesis, Leuphana University Lüneburg.

@mastersthesis{siu2026magictower,
  author = {Siu, King Wai},
  title  = {From Reactive Control to Holistic Planning: Evolutionary Prompt Optimization for Hybrid Deterministic-Neural Agents in Long-Horizon Environments},
  school = {Leuphana University Lüneburg},
  type   = {Master's thesis (M.Sc. Data Science)},
  year   = {2026},
  url    = {https://github.com/thetobysiu/mt-thesis}
}