Build Daily

Tinley Park · October 13, 2026
The series · 8 parts

Local Azure AI Development

Microsoft's on-device AI stack from the developer's seat — Foundry Local, the ONNX runtime, Olive, and the C# SDK. What it takes to run a model on the workstation instead of sending source to a lab, and which Microsoft credentials actually map to the work.

Start reading → Foundry Local vs Ollama for a .NET shop

  1. Foundry Local vs Ollama for a .NET shop

    Both run a language model on local hardware. The similarity ends there. One is a developer's workstation tool; the other is a native library that ships inside an application to an end user's device. For a .NET team, that distinction decides which one is even a candidate.

  2. What the ONNX runtime does to a model

    Foundry Local is a thin wrapper around a much older, much larger piece of engineering. Understanding what ONNX Runtime does — graph partitioning, execution providers, and a fallback that always exists — explains most of the surprises an on-device deployment produces.

  3. Olive and the compile path

    The catalog is not the boundary. Any Hugging Face model can be converted, quantized, and run locally — but the conversion is a real build step with real choices in it, and the documentation is unusually honest about how often a single command line fails to generalize.

  4. The Foundry Local C# SDK

    Two NuGet packages, one API surface, and a model lifecycle with four verbs. What it actually takes to put a language model inside a .NET application without a service to operate, a port to negotiate, or an installer for the customer to run.

  5. Where the model runs

    One brand, three deployment targets, and a set of names one word apart from each other. Sorting out the device runtime, the on-premises Kubernetes platform, and the cloud service — and the exact point where local AI stops being local and becomes self-hosted.

  6. How a model gets into a curated catalog

    A model trending on a public hub on Monday is not therefore runnable in a curated on-device catalog on Tuesday. The lag has a mechanism, the mechanism is public, and understanding it is the difference between waiting and contributing.

  7. The Microsoft credential map

    The role-based certification lineup is being rebuilt around AI, and several credentials that anchored career advice for years have already retired this summer. What the current map looks like, what each tier is actually for, and the one test worth applying before booking any exam.

  8. Becoming a Microsoft MVP

    There is no application form. The award cannot be requested, bought, or earned by passing anything — someone else has to put a name forward, and the evidence they point at is a year of public work that already happened. How the mechanism actually works, and what follows from it.