Master’s Thesis Presentation • Software Engineering • Facilitating Game Development From Requirements to Code with LLMs

Please note: This master’s thesis presentation will take place in DC 2310.

Ahmed El Shatshat, Master’s candidate
David R. Cheriton School of Computer Science

Supervisors: Professors Derek Rayside, Daniel Berry

Large Language Models (LLMs) are seeing further integration into the software development process, in part due to their strength as code-drivers. In defiance of prompt engineering, I put forth Software Requirements Specification (SRS) as the only necessary input to such systems. An extension of Bingyang Wei’s work on a progressive prompting method to develop code, this thesis explores the capacity of Wei’s Progressive Prompting Method (WPPM) to facilitate game development, using a Tailored Large-Language Model (TLLM). I investigate the following three research questions:

1. Does the WPPM facilitate game development?
2. What strengths and weaknesses emerge in the use of the WPPM to develop a game app?
3. What impacts do the results of such an exploration have on the software development process?

Through case studies exploring the main facets of game development, insights are extracted demonstrating that game development under such conditions is possible, and deserves further analysis against traditional development processes. This thesis highlights the strengths of LLMs for software development, and addresses how oversight by a domain expert can mitigate the associated weaknesses. This thesis also serves to combat dominant narratives over-estimating the capabilities of LLMs, which seek to push such systems as a silver bullet for all software problems. Further research in accurately assessing the capacity of the technology is suggested, highlighting the importance of a cost analysis against the traditional manual development process.