Public Institution Customer Service Assistant Service Powered by Hyper-scale AI

Overview

As the demand for non-face-to-face services grows, the workload and labor intensity of public call centers have significantly increased. To address this and improve public satisfaction, the project introduced an AI-powered Civil Complaint Assistant service utilizing Hyper-scale AI technology. By providing precise answers to inquiries along with relevant source information, this AI assistant service has successfully enhanced both citizen satisfaction and the operational productivity of consultants.

Project description

By integrating Hyper-CLOVA X with cloud-based search capabilities, this system enables the retrieval of publicly available civil complaint data and the identification of relevant search results on a large scale. The system provides consultants with automated summaries and source references via Generative AI, enabling them to verify answers quickly and significantly reduce response times.

To enhance response quality, the project trained the model on a vast dataset of unstructured administrative manuals and forms—including 834 cases from Suncheon City, 444 from Uijeongbu City, and 422 from Hwaseong City. By applying RAG (Retrieval-Augmented Generation), the service successfully mitigated hallucinations and improved the overall accuracy of the AI assistant.

For continuous service optimization, an Agile methodology was adopted to consistently refine answer relevance and user convenience. Furthermore, to objectively verify the improved accuracy of civil complaint responses, "Generative Q&A Model Accuracy" was established as a quantitative performance metric, with formal testing conducted by an accredited certification testing agency.

Key technological features

• NLP & API Integration: Processing natural language queries from consultants by performing keyword decomposition and synonym expansion, followed by seamless integration with the Hyper-CLOVA X API
• Accuracy Ranking Optimization: Tuning search result rankings to ensure high relevance and precise alignment with the user's intent
• Fine-tuning Data Construction: Building specialized datasets for model fine-tuning to accurately capture and reflect the specific information used by consultants
• Kubernetes Rolling Deployment: Establishing a CI/CD pipeline where development updates are continuously containerized and deployed to the cluster via Kubernetes rolling updates

Expected outcomes

The use of the AI Civil Complaint Assistant is expected to reduce the data search time for each inquiry by over 30 seconds, leading to a more than 10% increase in total call handling capacity. Additionally, it is anticipated that the daily workload of call center staff will be reduced by approximately 42 minutes, allowing for a 20% increase in their rest periods. Enhanced operational productivity for call center workers also signifies an improvement in citizen satisfaction. High-quality consultation is expected through faster processing and responses that consistently apply a professional tone and manner.

Based on this project, there are plans for an initial expansion to the three pilot call centers in Hwaseong, Suncheon, and Uijeongbu. Subsequently, the service will be continuously scaled to include nationwide public inquiry agencies, as well as private sectors such as finance, insurance, and logistics.

Launch year: 2023

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=29194