AI Transformation in Municipalities: Architecture and Data Management

At X Mind Solutions, we participated in the closing session of the “Research Project on AI Transformation for Municipalities.” At the workshop hosted by the Marmara Municipalities Union, we drew on our field experience to share practical perspectives on how municipalities should approach AI projects. Our contribution focused on distinguishing between project types, sharing computing resources and making architectural decisions that ensure data is kept in the appropriate environment.

One point we particularly emphasized during the session was that not all AI projects should be assessed within the same framework. Operational tasks based on large language models, analytical projects using machine learning and field automation should be classified separately. This distinction matters not only for identifying the technology used, but also for accurately assessing each task’s data needs, operating approach and infrastructure requirements. When planning transformation at the municipal level, the starting point should be to identify the nature of the need before selecting a model.

On the infrastructure side, we highlighted the potential cost advantages of shared GPU resources. With large language model usage concentrated during the day, allocating nighttime hours to running machine learning models could help balance resource utilization. The approach here is to plan workloads together rather than build separate capacity for each project. Its feasibility should be assessed based on project operating schedules, capacity requirements and suitability for sharing resources.

Regarding data management and Türkiye’s Personal Data Protection Law (KVKK), we highlighted the importance of local large language models. Where municipal data is processed and which systems it is transferred to are issues that need to be addressed at the outset of architectural design. We consider the use of local models a critical requirement for maintaining control over sensitive data. However, running a model in-house does not, on its own, ensure KVKK compliance; access permissions, data processing procedures and security measures must also be assessed together.

We do not see AI transformation in municipalities as merely a matter of model selection. Establishing the right architecture, keeping data in an appropriate environment and aligning solutions with operations on the ground must be addressed together. The approach we shared at the closing session was based on this integrated perspective. We thank the Marmara Municipalities Union for hosting the workshop and everyone who contributed, and we hope this work will serve as a reference point for municipalities when evaluating their AI projects.