Stop spending tonight copy-pasting from MS Learn. CSA-Copilot turns a single prompt into slide decks, demo guides, hackathon packages, and full Azure projects — researched, built, and QA-checked by 27 AI agents. You describe what you need. They deliver. You review and present.
You need a GitHub Copilot subscription and the GitHub CLI (gh auth login).
git clone https://github.com/olivomarco/vbd-copilot.git
cd vbd-copilot
./scripts/setup.sh # one-time: venv + frontend deps
./scripts/dev.sh # Browser UI → http://localhost:5173
./scripts/dev.sh --electron # Desktop UI
./scripts/dev.sh --cli # CLI (Docker)Note
Windows users: see docs/INSTALLATION.md for Windows-specific Docker commands ($(gh auth token) doesn't work in CMD).
Tip
setup.sh checks prerequisites (gh, uv, node) and installs everything. Pass --with-system-deps to also install LibreOffice and Poppler (PPTX thumbnails).
Warning
The CLI is battle-tested. The Browser / Electron UI is experimental — if something breaks, fall back to --cli.
More install options (Codespaces, Copilot plugin, native Python): docs/INSTALLATION.md
Four workflows. Each one researches official sources, asks for your approval at checkpoints, and runs automated QA before delivery.
| Workflow | What you say | What you get | Time saved |
|---|---|---|---|
| Presentations | "1-hour L300 deck on GitHub Copilot extensions" | .pptx + speaker notes + generator script |
4-6h → ~1h |
| Demos | "3 L300 demos on Azure Container Apps" | Step-by-step guide + runnable scripts + troubleshooting table | 3-4h → ~30min |
| AI Projects | "Automate claim processing for insurance" | Brainstorm → architecture → Bicep + app code + CI/CD + tests | Days → ~1h |
| Hackathons | "Full-day L300 hackathon on Container Apps" | Challenges + coach solutions + dev container + scoring rubric | 1-2 days → ~1h |
Important
This is deep research, not instant generation. A deck takes ~1 hour — but that replaces 4-8 hours of manual work. Kick it off and do something else.
Launchpad — pick your workflow |
Library — browse and preview outputs |
AI agents ask clarifying questions |
Full architecture diagrams, auto-generated |
Hackathon packages ready for Codespaces |
CLI — same power, no browser needed |
 |
 |
 |
 |
 |
| Title slide | Section divider | Content slide | Deep dive | Architecture |
Browse all samples: slides · demos · hackathons · AI projects
| Level | Audience | What it looks like |
|---|---|---|
| L100 | Executives | Value props, no code |
| L200 | Technical decision makers | Architecture, key concepts |
| L300 | Practitioners | Code samples, implementation patterns |
| L400 | Experts | Internals, performance tuning |
"Create a 15min L200 briefing on what's new in AKS"
"Build a 30min deck from notes/aks-security-review.md"
"@ai-brainstorming AI use cases for a healthcare company"
"@ai-solution-architect Design architecture for idea #3"
"@ai-implementor Implement the solution"
"@hackathon-conductor 4-hour L300 hackathon on AKS"
/resume # pick up where you left off
/model # switch models mid-session
Every output passes through multiple checks before delivery. Details: docs/QUALITY.md
- Research restricted to official sources (MS Learn, docs.github.com, devblogs, techcommunity)
- Human approval stops before builds and before delivery
- Automated QA validators for PPTX, architecture, infra, pipelines, docs, and hackathons
- Content humanization — AI-tell detection with automatic rewrites
- 4-reviewer gate on AI projects (code, infra, pipeline, docs must all approve)
- 80% test coverage enforced via
pytest --cov