Creating Business Value
with AI Solutions

Problem

Payment and operational data from regional currencies across multiple municipalities were fragmented, making data-driven policy decisions difficult.

Solutions

• ▹ Enterprise Data Lakehouse: Designed unified Data Lake collecting from heterogeneous DBs (Oracle, MySQL) across municipalities
• ▹ ETL Pipeline Automation: Apache Spark-based batch processing system handling millions of daily transactions
• ▹ Data Governance: Implemented RBAC-based access control and pseudonymization for sensitive financial/personal data

Problem

Customer inquiries for 9 municipalities' regional currency services surged, requiring CS cost reduction and 24/365 response capability.

Solutions

• ▹ OSMU Engine: Multi-Tenant architecture with single AI core handling 9 different municipal policies
• ▹ Hybrid NLP Model: Rule-based + KoBERT/BERT Intent Classification for maximum accuracy
• ▹ MLOps Pipeline: Continuous model improvement through automatic retraining with conversation history

Problem

Employees' repetitive questions about internal policies, IT support, and HR needed automated responses. MS Teams integration was essential.

Solutions

• ▹ MS Teams Bot: Instantly usable within Teams without separate app installation
• ▹ RAG Model: Vector-based search and answer generation from internal wiki (Confluence) and policy documents, minimizing hallucination
• ▹ Security Protocol: Hybrid On-Premise + Private Cloud configuration compliant with enterprise security guidelines

Problem

Beyond simple payment, a customized benefit system was needed to boost local merchant sales and enhance user benefits.

Solutions

• ▹ Hybrid Recommendation: Collaborative Filtering + Content-Based Filtering combination for improved accuracy
• ▹ Real-time Context: "Places to visit now" recommendations combining user location (Geo-fencing), weather, and time
• ▹ User Profiling: Vector-based similar merchant recommendations using activity area, preferred categories, and consumption time

Problem

Tens of thousands of business registration certificates and bank statements submitted for merchant registration were being manually reviewed and entered, causing inefficiency.

Solutions

• ▹ Image Classification: CNN-based model (EfficientNet) for automatic document type classification
• ▹ OCR Extraction: Clova OCR + Tesseract ensemble for key field (business ID, account number) text extraction
• ▹ Verification: Automatic validity check through National Tax Service and Financial Settlement API integration
• ▹ HITL Interface: UI/UX for human review of only low-confidence AI predictions