Zava AI MicroHack — Build Enterprise AI Agents on Microsoft Azure

A 1-day, instructor-led, Train-the-Trainer microhack that teaches partners to design, build, and govern a multi-agent AI application on Microsoft Azure — and produces audit evidence for the AI Applications on Microsoft Azure Advanced Specialization (Module B) along the way.

---

🎯 What you build

You are the platform team at Zava, a fictional intelligent athletic apparel retailer. Over one day you build the Zava AI Agent Platform: a set of cooperating enterprise agents (customer concierge, merchandiser, operations, and a trust & compliance agent) on Microsoft Agent Framework, grounded on Zava data, deployed to Azure, observable, and governed.

flowchart LR
  U([User / Store Associate]) --> ORCH[Orchestrator Agent]
  ORCH --> CON[Concierge]
  ORCH --> MER[Merchandiser]
  ORCH --> OPS[Operations]
  ORCH --> TRU[Trust & Compliance]
  subgraph AI[AI Services pillar]
    FND[Azure AI Foundry / OpenAI]
    SRCH[Azure AI Search]
    DOC[Document Intelligence]
    CS[Content Safety]
  end
  subgraph DATA[Data pillar]
    COS[(Cosmos DB)]
    PG[(PostgreSQL + pgvector)]
  end
  subgraph APP[App Platform pillar]
    ACA[Azure Container Apps]
    APIM[API Management]
  end
  CON & MER & OPS & TRU --> FND
  CON --> SRCH & COS
  MER --> PG & COS
  OPS --> DOC & COS
  TRU --> CS
  ACA --- ORCH
  APIM --- ACA

Loading

Modular inference: every AI call runs on GitHub Models (free, default), Azure OpenAI, or Azure AI Foundry — switched by a single environment variable. Early labs cost $0; later labs azd up into a real Azure subscription.

🧑‍🏫 Two tracks, one story

Track	Stack
🐍 Python	FastAPI + agent-framework (Microsoft Agent Framework)
🟦 .NET (C#)	ASP.NET Minimal API + Microsoft.Agents.AI (Microsoft Agent Framework)

Pick a track — both implement the same agents, the same HTTP API, and the same labs.

🗺️ The 1-day agenda

See docs/agenda.md. Labs:

Lab	Title	Module B focus	Pillar
0	Environment & GitHub Models	—	—
1	Your first Zava agent (modular inference)	B.1.1	AI
2	Ground it: RAG with AI Search + pgvector	B.1.1	AI + Data
3	Deploy to Azure: Container Apps via azd	B.1.1	App Platform
4	Data agents: Cosmos DB catalog & orders	B.1.1	Data
5	Trust, safety & observability	B.4.1	AI
6	Capstone: GitHub agents, skills & MCP + evidence pack	B.1.1–B.4.2	All

🤖 GitHub agents, skills & MCP

This repo ships GitHub Copilot custom agents, skills, and MCP servers (including a custom Zava MCP server) so learners extend the platform with AI. See .github/agents/, .github/skills/, and mcp.json.

🏅 Module B / Advanced Specialization mapping

Every lab output is also audit evidence. See docs/module-b-mapping.md and docs/cert-crosswalk.md. Source controls: NikoMix/specialization-ai-apps.

🚀 Quick start

# 1. Open in the dev container (VS Code: "Reopen in Container") — Python, .NET, azd, Bicep preinstalled
# 2. Run an agent on GitHub Models (free):
export GITHUB_TOKEN=<your PAT with models:read>
#   Python:
cd src/python && make run
#   .NET:
cd src/dotnet && dotnet run
# 3. Later labs — provision Azure:
azd up

📁 Repository structure

Path	Contents
docs/	Overview, agenda, architecture, facilitator guide, labs, solutions, evidence templates
slides/	Marp decks (export to HTML/PDF/PPTX)
src/python · src/dotnet	The two reference tracks
src/mcp-zava	Custom Zava MCP server
infra/ · azure.yaml	azd + Bicep (Azure Verified Modules)
data/	Zava seed data
.github/	Copilot agents, skills, instructions, workflows
.microhack/build-contract.md	The internal build specification

👩‍🏫 For trainers

Start with the Facilitator Guide — timing, talk tracks, "how to teach this", common pitfalls, and reset/troubleshooting steps.

📄 License

MIT. Built with open-source tooling only — no paid licenses required.