Education · Analytics

The Power BI Fellowship

Most Power BI courses teach features. This one teaches the job: the vague requests, disputed numbers, failed refreshes and design decisions that BI developers usually meet only after years at work.

Type
Open-source learning platform
Content
3 levels · 7 tracks · 18 scenarios · 30 guides
Runs on
GitHub Pages; progress stays in the browser
License
MIT
The Power BI Fellowship home page: “Learn Power BI by doing the work.” above three paths: learn the skills, practise real work and prepare for interviews.

Overview

The Fellowship is organized as Learn → Practise → Debug → Ship → Operate → Architect. It needs no account and no server, and it tracks nothing by default.

  • Skill Mode: three levels and seven tracks (SQL for BI, warehousing, testing, automation and APIs, governance, Fabric and modern Power BI). Each assignment has an expected result you can check and a worked solution.
  • Experience Mode: ten connected sprints and eight drills at Northwind Outdoors, a fictional retailer, with emails, chat messages, evidence files, progressive hints, a rubric and a model answer.
  • BI Developer Toolkit: one search across 30 guides, including “what’s broken?” decision trees, a performance clinic, a production runbook and an error decoder, plus a library of templates and datasets.

Motivation

I spent several years building and administering Power BI reports. Most of the hard parts of that job are not features. They are judgment calls: reconciling a number Finance disputes, finding why a page takes 18 seconds, deciding between a hotfix and a rollback. The Fellowship hands you the evidence and makes you decide.

How it’s built

An Experience Mode incident ticket, INC-2044: “The executive dashboard takes 18 seconds”, reported by the fictional CEO, with priority, a five-hour time box, and a Take the ticket button.
An Experience Mode incident. Labels read the way a reporter writes them and never name the cause.
  • Content lives as JSON in content/, and a build step in tools/ generates the static site.
  • Pages are readable without JavaScript, and the build fails if a hint, rubric or model answer leaks into them.
  • Any scenario can be made available offline.

Tested, not trusted

  • Every dataset is generated from a fixed seed, and its counts, nulls, duplicates, keys and planted defects are asserted. Expected results are computed from that data at build time.
  • SQL answers are executed on SQLite and on SQL Server 2022, and both engines must return the same rows.
  • Critical journeys, accessibility (axe-core, both themes) and layout at four widths are tested in a real browser on every push. Nothing deploys unless everything passes.
  • Fast-moving content carries a verification date, and an issue opens automatically when its review is due.

Progress you can trust

Progress is shown separately as verified (your first quiz answer), self-assessed and recall (spaced repetition), and each figure explains how it is calculated. Professional stages, from Data Analyst to Architect, are cleared by finishing well-rated scenarios, not by ticking boxes.

Contact

Have something interesting in mind?

I’m always interested in thoughtful collaborations, interesting research problems, and opportunities to build useful things.