|
Large language models for TOR-based specification compliance checking in public procurement |
|---|---|
| รหัสดีโอไอ | |
| Title | Large language models for TOR-based specification compliance checking in public procurement |
| Creator | Supaphit Worakijtanait |
| Contributor | Warut Pannakkong, Advisor |
| Publisher | Thammasat University |
| Publication Year | 2568 |
| Keyword | TOR compliance, Public procurement, Large language model, GPT-5, AI-assisted evaluation, Prompt engineering, LLM specification comparison |
| Abstract | In Thailand’s public procurement system, verifying whether products meet the required Terms of Reference (TOR) is a recurring challenge, especially in technology-related projects. TOR documents issued by the Comptroller General’s Department contain detailed technical and financial requirements that must be strictly followed to ensure fairness and transparency. In practice, engineers must manually review and match every TOR condition against the specifications of proposed products, a process that requires substantial time, human effort, and technical expertise.This study aims to design a prototype framework that applies Large Language Model (LLM) to assist in TOR compliance checking. Instead of using complete TOR documents, selected sections from real government TORs were extracted and used as representative samples for model testing. These TORs were taken from official Thai procurement cases to ensure neutrality, consistency, and alignment with national standards. GPT-5 was employed as an analytical model to interpret the requirements and perform structured comparisons between TOR clauses and product specifications across three product categories: notebooks, desktop computers, and servers.This study presents the prototype as an initial exploration of how AI-assisted evaluation could be integrated into government procurement workflows. The proposed framework provides a foundation for future research and system development aimed at enhancing efficiency, objectivity, and reproducibility in TOR evaluation. |
