How azure-samples/js-ai-build-a-thon works
Project type and purpose
• This repository is not a traditional full-stack application but a “template” / tutorial orchestration system. It uses GitHub as the platform, GitHub Actions for CI/CD workflows, and Azure AI services (via the Azure AI Foundry & AI Toolkit extensions) to teach developers how to build and deploy JavaScript/TypeScript AI applications step by step.
File structure
• Top-level:
• .github/steps/0-10-*.md – sequential learning modules
• .github/workflows/*.yml – the pipelines that correspond to each step
• README.md – dynamically updated with instructions and progress
• Key patterns:
• Orchestration: each workflow triggers on a README change or on a push, updates the README, and moves the student to the next module.
• Template repository: developers instantiate their own copy (via “Start Course”), gaining a public repo with these files.
• Azure integration points: later steps connect to Azure AI Foundry, LangChain.js, RAG, and build an AI agent.
README for additional insights
• The README describes the learning journey:
1. Local AI prototype (GitHub Models + AI Toolkit)
2. Migrate to Azure AI Foundry
3. Deploy and code against Azure
4. RAG (LangChain.js)
5. Build & extend an AI agent
6. Automate with GenAIScript
Main components
a. Developer workstation
• Local code editor
• AI Toolkit + Azure AI Foundry VSCode extensions
b. GitHub
• Template repository
• Developer’s forked repository
• GitHub Actions (10-step pipelines)
c. Azure Cloud
• Azure AI Foundry (OpenAI models, embeddings, APIs)
• Optional: Azure Cognitive Search (for RAG)
d. External libraries/services
• GitHub Models & GenAIScript
• LangChain.js
Relationships and data flow
• Developer clicks “Start Course” → GitHub new repo from template
• Push to GitHub → GitHub Actions step0 → updates README with step1
• Developer follows instructions locally (AI Toolkit runs, code changes) → commit/push
→ GitHub Actions step1 → scaffolds Azure Foundry connections → update README to step2
• Repeat until final step → full AI agent deployed to Azure
• In parallel: LangChain.js + Azure Cognitive Search integrate for RAG, agent built with MCP tools
Architectural patterns & design principles
• Pipeline-as-Code: each learning step is automated by a CI workflow
• GitOps style: the state of the tutorial is encoded in the README and repository state
• Modularity: separate workflows, separate steps, clear progression
• Extension points: students can replace templates, adjust Azure resource definitions