ACM SAC 2026 Conference paper

Requirement Bot: Enhancing Software User Requirement List Through Retrieval-Augmented Human-AI Co-Authoring

Wangfan Li, Carlos Toxtli-Hernández

Proceedings of the 41st ACM/SIGAPP Symposium on Applied Computing, 2026 · pp. 1437-1443

Abstract

Poorly defined user requirements can greatly contribute to cost overruns and even project failures in software development. With the recent advancements in large language models (LLMs), it might be possible to utilize them during the requirement engineering process. In this study, we present Requirement Bot, a RAG-backed assistant designed to support the creation of software user requirement lists in live human-AI collaboration. For system evaluation, we asked 54 participants with software engineering experience to each create requirements for an online book lending library application and test across four conditions: human-only, human-AI with base GPT-4.1, human-AI with Requirement Bot powered by GPT-4.1 and RAG, and an AI-only approach using GPT-4.1. Requirement lists were evaluated by 32 experienced software developers using standardized criteria (unambiguous, necessary, consistent, verifiable, feasible, complete, and understandable). Results show that human-AI collaboration using Requirement Bot significantly outperformed all other conditions in received scores, showing significant improvement over both base AI collaboration and purely human-created lists. AI-involved conditions also significantly outperformed the human-only condition, particularly when it comes to completeness and understandability of the requirement list. By providing domain-specific knowledge through Retrieval-Augmented Generation, it substantially improves human-AI collaborative requirement generation by increasing completeness. Future research should focus on detailed content analysis, measure human effort as a factor, and evaluate AI-assisted requirement generation in real-world deployments.

Cite this work

Wangfan Li and Carlos Toxtli-Hernández. 2026. Requirement Bot: Enhancing Software User Requirement List Through Retrieval-Augmented Human-AI Co-Authoring. Proceedings of the 41st ACM/SIGAPP Symposium on Applied Computing. https://doi.org/10.1145/3748522.3780008

@inproceedings{Li_2026b,
  series = {SAC ’26},
  title = {Requirement Bot: Enhancing Software User Requirement List Through Retrieval-Augmented Human-AI Co-Authoring},
  url = {http://dx.doi.org/10.1145/3748522.3780008},
  doi = {10.1145/3748522.3780008},
  booktitle = {Proceedings of the 41st ACM/SIGAPP Symposium on Applied Computing},
  publisher = {ACM},
  author = {Li, Wangfan and Toxtli, Carlos},
  year = {2026},
  month = Mar,
  pages = {1437--1443},
  collection = {SAC ’26}
}