Urban planning increasingly relies on 3D city models and geospatial databases for sustainable development. However, no complete system currently offers immersive 3D visualization of building attributes in augmented reality (AR), bidirectional attribute editing propagated back to an authoritative geodatabase, and real-time synchronization across multiple clients. In this paper we present a full-stack, multi-level architecture that bridges Microsoft HoloLens 2 AR visualization with a CityGML/3DCityDB geospatial backend via a Django middleware layer. A hybrid WebSocket+HTTP polling communication layer, built on Django Channels with Redis pub/sub and a platform-abstracted Unity client, delivers sub second server-push updates (200-500ms latency) while automatically falling back to HTTP polling when WebSocket connections are unavailable, reducing per-client bandwidth by over three orders of magnitude compared to periodic polling. The architecture has been validated on 5,005 buildings in the German municipality of Bisingen. For future research and development plans, it would be valuable to focus on geometry editing workflows and the integration of AI-driven interaction. In particular, enabling voice-based commands would significantly improve usability on HoloLens 2, where typing is inconvenient for most users.