A growing series of 3-4 hr, self-guided technical workshops that teach you how to ship production-grade AI applications on Microsoft Foundry — one model family at a time. Workshops target L200–L400 learners (comfortable with Python, terminals, and cloud basics) and are designed for both live delivery and asynchronous self-paced study.
Each workshop is opinionated about the same end-to-end arc so the muscle memory transfers from one model family to the next:
| Phase | Description |
|---|---|
| Setup | Provision a Foundry project, deploy the models the workshop needs, and verify your environment from GitHub Codespaces. |
| Build | Implement a realistic scenario (often an agent) using the model family's flagship capabilities through Foundry's SDKs. |
| Explore | Go deeper on the model family's specialised capabilities (embeddings, rerank, tool use, long context, vision, etc.) and connect them back to the scenario. |
| # | Workshop | Model family | Status | Start here |
|---|---|---|---|---|
| 1 | Anthropic on Foundry | Claude family | ☑️ WIP | anthropic/ |
| 2 | Cohere on Foundry | Command A, Embed v4, Rerank v4 | ✅ Ready | cohere/README.md |
| 3 | Intro To Foundry | Cross-model platform features | ☑️ WIP | foundry/ |
Each workshop is fully self-contained. You can start with whichever model family you care about — there are no cross-workshop prerequisites.
Every workshop follows the same top-level shape so you always know where to look:
<model-family>/
├── README.md # workshop overview — start here
├── sample.env # canonical environment-variable reference
├── requirements.txt # Python dependencies for the workshop
├── lab-0-setup/ # provisioning + env wiring (Codespaces-first)
├── lab-1-<scenario>/ # build a realistic application with the flagship model
└── lab-2-<deep-dive>/ # explore the model family's specialised capabilities
Open the workshop's README.md first — it links to lab-0-setup/SETUP.md, which walks you through provisioning in roughly 15–20 minutes.
📝 Environment variables. Every workshop ships a
sample.envin its top-level folder that documents every variable the labs read (Foundry endpoint, deployment names, API keys, optional load-test knobs, etc.). Lab 0'ssetup-env.shcopies it to.envand fills in concrete values for you. Readsample.envfirst whenever you want to know what a workshop expects to find in its environment.
- An Azure subscription with permission to create a Foundry account, project, and model deployments.
- GitHub Codespaces (recommended) or a local environment with Python 3.11+ and the Azure CLI.
- Familiarity with Jupyter notebooks and the Python ecosystem.
Each workshop's lab-0-setup/SETUP.md lists any model-family-specific extras (regions, quotas, SDK versions).
The workshops live side-by-side so improvements to one (notebook style, evaluator patterns, tracing setup, Codespaces hardening) can flow easily to the others. If you spot a gap or an inconsistency between workshops, open an issue or PR against the affected <model-family>/ folder.