e-RFP (Request for Proposal) Assistant Service

Overview

The Public Procurement Service (PPS) aims to develop a service that processes and learns from 25,000 Request for Proposal (RFP) documents (totaling 200 million tokens) registered in its system. This service enables the automated generation of RFP drafts organized by major table of contents. Building upon the Proof of Concept (PoC) conducted in 2023, the project focuses on advancing and optimizing the draft generation model for practical administrative use.

Project description

This service supports the generation of RFP drafts by collecting and processing data based on major sections and chapters, which are then used to train a Large Language Model (LLM). To respond effectively to various types of service RFPs, the team implemented LoRA to train and compare multiple model iterations efficiently. This lightweight approach ensures the system can find the optimal model structure without excessive computational costs.

Furthermore, to improve the quality of training data and the accuracy of generated drafts, the system leverages GPT-4’s natural language understanding capabilities to automatically extract core keywords. By incorporating a user-validation stage, the platform ensures the validity of these keywords before they are used for generation, thereby significantly enhancing data precision. As a result, users receive rapid and precise support in drafting RFPs for specific service projects, benefiting from a high-performance automated service that ensures both productivity and consistency in document preparation.

Key technological features

• LoRA (Low-Rank Adaptation): Application of LoRA, a PEFT (Parameter-Efficient Fine-Tuning) technique, to enable efficient training and testing of multiple model versions in a lightweight manner
• Keyword-driven Precision Enhancement: Utilizing GPT-4 for automated keyword extraction, followed by a manual verification step by users to maximize data precision and relevance

Expected outcomes

This service is expected to significantly assist demanding organizations in the complex task of drafting RFPs. By streamlining the initial writing phase, it will lower the barrier to document creation and contribute to the revitalization of the "e-RFP Helper System." Ultimately, the automation of specialized document drafting will reduce administrative lead times, minimize human error, and standardize the quality of public procurement documents across various government sectors.

Launch year: 2024

Attribution

Case study courtesy of the National Information Society Agency, Republic of Korea, Apolitical's Content Partner. Sourced from Use Cases of Public AI Service, Vol.2, https://eng.nia.or.kr/site/nia_eng/ex/bbs/View.do?cbIdx=31975&bcIdx=29182&parentSeq=29231