Challenge
Data integration was limited to manual local file uploads, making it difficult to connect diverse internal and external sources and reducing usability for domain experts. Limited library support, siloed development environments, and manual model deployment disrupted the machine learning lifecycle. As a result, models were difficult to reuse and often ended up as one-off implementations.
Approach
Internal and external sources—such as data portals, servers, and national information systems—were integrated via APIs. A user-friendly, UI-based workflow was introduced, enabling users to independently develop and deploy models and improving overall accessibility. For the first time in the public sector, an AI operations system (MLOps) was deployed in a Kubernetes (K8s) environment compliant with National Intelligence Service security standards, enabling secure CI/CD/CT across the ML lifecycle.
Value Delivered
Access to a broader range of data sources enabled more structured data management and enhanced the flexibility and reusability of AI model development. The intuitive interface allowed users to independently deploy and monitor models, improving the reliability, maintainability, and security of AI operations in public sector environments.
