Build production-ready applications from a product specification, deployed to Azure. Two phases: Discovery (understand what to build) and Delivery (build and ship it).
Pick a shell template that matches your tech stack. The shell provides project structure, test frameworks, Azure infrastructure templates, and CI/CD workflows.
| Template | Language | Stack |
|---|---|---|
| Next.js | TypeScript | React + Express + PostgreSQL |
| .NET | C# | ASP.NET Core + Entity Framework |
| Agentic .NET | C# | ASP.NET Core + Claude AI Agent SDK |
| Agentic Python | Python | FastAPI + Claude AI Agent SDK |
Your PRD is reviewed through product and technical lenses. Up to 5 refinement passes ensure completeness: edge cases, feasibility, clarity. The PRD is then broken into individual FRDs (Feature Requirement Documents).
Human Gate: Review and approve refined PRD and FRDs before proceeding.
Interactive HTML prototypes are generated from the approved specs and served via HTTP for review.
Human Gate: Review prototypes, provide feedback, approve designs.
FRDs are organized into ordered delivery increments. The first increment is always a "walking skeleton"—the minimal end-to-end path that proves the architecture works.
Every technology choice is researched, evaluated, and documented as an Architecture Decision Record (ADR).
Human Gate: Review tech stack choices and ADRs.
- E2E specs (Playwright) from UI flows
- Gherkin scenarios from FRDs
- Unit tests (Vitest) scaffolded
- All tests run and establish a "red baseline" (they fail, as expected)
Human Gate: Review and approve Gherkin scenarios.
- OpenAPI specs define API boundaries
- Shared TypeScript types for request/response DTOs
- Infrastructure contracts (Azure resources via Bicep)
- Contracts enable parallel implementation of API and Web slices
Three parallel slices:
- API Slice: Express routes, services, models (tested with Vitest + Supertest)
- Web Slice: Next.js pages, components (tested with builds + component tests)
- Integration Slice: Wire API + Web together (tested with Cucumber + Playwright e2e)
API and Web run in parallel. Integration is sequential after both complete. Red baseline tests turn green.
Human Gate: PR review before deployment.
- Azure deployment via azd + Bicep
- Smoke tests: health checks, API health, frontend load, critical user paths
- Full e2e regression against deployed URL
- Rollback + GitHub issue if smoke tests fail
After Step 4 completes, loop back to Step 1 for the next increment.
Every increment ships with:
- ✓ 100% test coverage (unit, integration, e2e)
- ✓ All tests passing on main
- ✓ Live deployment to Azure
- ✓ Rollback capability on failure
- ✓ Zero manual deployment steps