How This Was Built

How this portal was built

No GUI was opened. No drag-and-drop. No scheduled refresh wizard.

This entire portal was generated from:

SQL queries against a Microsoft Fabric Lakehouse
Markdown files in a Git repository
One npm run build command

The tool is Evidence.dev — open source, MIT licensed. The data lives in a Fabric F2 capacity on Azure Sponsorship credits. The build output is a static HTML site deployed to GitHub Pages via a GitHub Actions workflow on every git push.

The full stack

Layer	Technology	How it was built
Data capture	Power Platform Inventory API + CoE Dataverse export	Python scripts, PAC CLI
Storage	Fabric Lakehouse (Delta tables)	PySpark notebook, Fabric REST API
Transformation	PySpark notebooks (bronze → silver)	Claude Code CLI + Fabric MCP
Portal	Evidence.dev	Claude Code CLI / Codex
Hosting	GitHub Pages	`git push` → GitHub Actions

Total GUI interactions to build this portal: 0.

Example: The heatmap on the overview page

The governance heatmap is this SQL query in sources/pp_inventory/signals_by_environment.sql:

select
  environment_name,
  signal,
  count(*) as signal_count
from silver_governance_signals
group by environment_name, signal
order by environment_name, signal

And this Evidence markdown in pages/index.md:

That is the entire definition of the chart. The SQL runs once at build time against the Fabric Lakehouse SQL analytics endpoint. The output is a pre-rendered static HTML file that loads in milliseconds — no database connection required at read time.

Example: The CoE drift alert

The alert on the CoE environment page is not hardcoded. It is conditional Evidence markup that fires whenever total_drift > 0:

The 133-flow delta in the CoE environment triggers this automatically from the data. A different tenant, different numbers — same code.

The deployment pipeline

Every git push to main triggers this GitHub Actions workflow:

- run: npm run sources    # queries Fabric Lakehouse
- run: npm run build      # generates static HTML
- uses: peaceiris/actions-gh-pages@v4
  with:
    publish_dir: ./build  # pushes to gh-pages branch

Credentials are stored as GitHub Actions secrets — never in the repository. The static output has no connection strings. The Fabric SQL analytics endpoint requires a valid Entra token to query; the published site has no path back to it.

To rebuild with fresh data without a code change: Actions tab → "Deploy to GitHub Pages" → Run workflow.

The 2020 answer vs the 2026 answer

2020: Install CoE Starter Kit (multiple components, multiple days), wait for initial sync (24–48h), build Power BI reports manually, maintain a scheduled refresh, manage a data gateway, explain to users why the numbers don't match native inventory.

2026: Capture inventory via API (Python, 10 minutes), land in Fabric Lakehouse (notebook, 30 minutes), generate portal from SQL + markdown (Claude Code, 2 hours). Deploy with git push. One afternoon.

The underlying principle has not changed: Inventory is data. Treat it like data.

What changed is that the tooling finally caught up.

Source: github.com/jukkan/pp-inventory-portal