GENCITY operates on a three-layer architecture: Azure manages policy and lifecycle, the orchestration layer coordinates workloads, and edge nodes execute AI locally. No raw data traverses the boundary between local and cloud.
The architecture separates the control plane (Azure) from the execution plane (edge nodes). This separation ensures that sensitive data is never transmitted to cloud compute resources.
Azure provides centralized governance: fleet management, policy distribution, identity services, secure update delivery, and anonymized telemetry collection. Azure never receives or processes raw operational data from edge nodes.
This middle layer routes workloads to available edge nodes, manages model versions, schedules inference tasks, and applies hardware-assisted anonymization before any data is logged or reported upstream.
Physical edge nodes run AI models locally. Each node contains an inference runtime, local model storage, and secure boot attestation. Nodes operate independently during connectivity interruptions and synchronize metadata when connectivity resumes.
RESTful APIs enable integration with existing enterprise systems. The local API gateway provides authenticated access to inference endpoints without cloud routing.
Operational health, performance metrics, and workload telemetry are collected and reported to Azure — but only after hardware-level anonymization strips all sensitive content.
Deploy one node or hundreds. The orchestration layer distributes workloads across available nodes based on capacity, proximity, and policy. Nodes can span multiple physical sites.
Automatic failover ensures workload continuity if a node becomes unavailable. Lifecycle management — provisioning, updates, health checks, decommissioning — is handled through Azure.
The following table summarizes what data crosses each boundary in the GENCITY architecture.
| Boundary | Data Type | Direction | Raw Data Exposure |
|---|---|---|---|
| Edge → Orchestration | Inference results, node health | Upstream | Local only — no cloud |
| Orchestration → Azure | Anonymized telemetry, policy ACKs | Upstream | None — hardware anonymized |
| Azure → Orchestration | Policy updates, model manifests | Downstream | N/A — control metadata |
| Orchestration → Edge | Model deployments, inference tasks | Downstream | Local only — no cloud |
| Edge → Cloud (direct) | — | Blocked | No direct path exists |