REprompt Improves AI Code Generation Satisfaction
Nanyang Technological University and East China Normal University introduce REprompt, a research framework treating prompts as requirements specifications for AI code generation. In tests on a vibe-coding platform with human evaluators, REprompt achieved satisfaction scores of 6.3 out of 7 for games and 6.5 out of 7 for utility tools, outperforming naive prompting, zero-shot chain-of-thought, and MetaGPT. The approach reframes prompting as requirements engineering to improve software outputs.
Key Points
- 1Demonstrates REprompt achieves 6.3/7 for games and 6.5/7 for utility tools in evaluations
- 2Explains prompting as requirements engineering, yielding more complete, accurate software generation than naive prompts
- 3Advises practitioners to craft structured specifications instead of single-shot commands to improve model outputs
Scoring Rationale
Strong empirical results and actionable methodology, but limited publication detail and unclear generalization beyond the tested platform.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
