Improving Military Manpower Administration Public Services using Hyper-scale AI

Overview

The existing chatbot provided for military service consultations followed a rule-based structure, which lacked flexibility in responses and presented difficulties in data management and maintenance. To address these challenges, the Military Manpower Administration (MMA) aims to enhance the efficiency of its consultation services by introducing Hyper-scale AI, overcoming the technical and managerial limitations of the previous system.

Project description

This service integrates Hyper-scale AI into the existing rule-based civil complaint (text-based) chatbot of the Military Manpower Administration. First, a Vector DB was constructed using military service consultation manuals and chatbot glossaries to provide the foundation for the service. As a result of optimizing data through specialized preprocessing, the system achieved a response accuracy of 94% and offers multi-language consultation, including English, Vietnamese, and Indonesian.

Considering that overseas citizens are also eligible for enlistment; the multi-language service allows users to select their preferred language for inquiries. The AI understands natural language questions and provides answers based on official manuals and glossaries. The responses are personalized, presenting direct answers to questions alongside relevant similar case studies.

Key technological features

• Data Preprocessing: Type conversion, unstructured data refinement, and JSON transformation (includes agency-specific processing such as URL handling, synonym dictionary application, and subtitle separation)
• Model Selection & Configuration: Selected customized models including an Embedding model (in-house model based on E5, fine-tuned by SDS), a Reranking model (in-house model based on E5, fine-tuned by SDS), and an optimized Samsung LLM
• System Optimization: Achieved performance optimization through meticulous tuning of Parsing and Embedding parameters.

Expected outcomes

By providing a context-based conversational service rather than a simple menu-selection method, this project enhances user satisfaction. Furthermore, the automation of data preprocessing and DB conversion is expected to deliver significant achievements in terms of data management and maintenance efficiency.

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=29199