AI on the Edge is a year-long Azure and edge AI project focused on making local inference governable, observable, policy-driven, and repeatable.
The project uses a .NET gateway and model-router pattern so applications can call one OpenAI-compatible facade while requests route to local, edge, cloud, Azure Local, Tenstorrent-backed, or mock inference endpoints based on policy and health.
Tagline: Cloud-governed, locally executed.
Relationship to the Private Cloud Lab
This is a separate project from the Private Cloud Tenstorrent Multi-Cloud Edge Lab.
The private cloud project owns the physical lab, Tenstorrent/NVIDIA hardware, Kubernetes environment, storage, and multi-cloud edge infrastructure. AI on the Edge uses that lab as one execution environment and demo lane, but the AI platform pattern is broader than the lab itself.
Current Focus
- Build the AI on the Edge Gateway and routing policy engine.
- Add private RAG with local embeddings, citations, and local-only denial behavior.
- Add a camera fleet operations assistant over the existing
cameras/#MQTT contract. - Add Azure Arc, GitOps, Azure Monitor, Managed Prometheus, Grafana, Key Vault, and policy evidence.
- Keep laptop-only demo mode deterministic with mock backends and reset controls.
- Treat Tenstorrent as an accelerator backend behind the same gateway contract.
Series
| Date | Post |
|---|---|
| Jul 13, 2026 | AI on the Edge: Local AI Without Local Chaos |
| Jul 20, 2026 | One App, Many Places to Run AI |
| Jul 27, 2026 | Private RAG That Cannot Leave the Edge |
| Aug 3, 2026 | From Camera Events to Operator Guidance |
| Aug 10, 2026 | Edge AI You Can Actually Operate |
| Aug 17, 2026 | When the Edge Has to Stand Alone |
| Aug 24, 2026 | Specialized Hardware Without an App Rewrite |
| Aug 31, 2026 | Building the AI on the Edge Demo System |
| Sep 7, 2026 | The AI on the Edge Reference Architecture |
Reference Workload
The reference workload is the camera fleet from the Azure IoT Operations series:
- Cameras on Azure IoT Operations, Part 1: The Control Plane and Network Model
- Cameras on Azure IoT Operations, Part 2: Swapping in the AIO MQTT Broker
- Cameras on Azure IoT Operations, Part 3: Data Flows, Connectors, and the Cloud
The AI project consumes that telemetry and runbook context, but the camera control plane remains its own prerequisite project.
Acceptance Criteria
The project is useful when a reader can understand the architecture, follow the demo build order, run the laptop demo, see where Azure governance fits, and know which existing posts cover the lab and camera prerequisites.