end-to-end-pbi-report-w-AI

KO | EN

gitlite — search

#javascript #react #ai #llm #golang #machine-learning #rust #docker #java #awesome #cli #mcp #openai #ios

end-to-end-pbi-report-w-AI

(★ 17)

Testing the end to end Power Bi report creation

end-to-end-pbi-report-w-AI Latest Version Download

Download Latest Version (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# LEGO Supplier OTIF Dashboard

A portfolio Power BI project built end-to-end using **Claude Code + MCP servers + domain-specific Skills** — no manual DAX authoring in the Desktop UI, no point-and-click report building.

The dashboard tracks supplier **On-Time-In-Full (OTIF)** performance for a fictional LEGO Supplier Services team, modelled on the kind of analysis a BI Manager Specialist in the Business Service Organization would own.

---

## What was built

| Layer | Artefact |
|---|---|
| Synthetic data | Python script generating ~18 months of PO delivery lines across 30 suppliers, 7 plants, 9 categories — with a deliberate anomaly injected in Oct 2025 |
| Semantic model | PBIP / TMDL star schema: `Fact_PODelivery`, `Dim_Supplier`, `Dim_Date`, `Dim_Plant`, `Dim_MaterialCategory` |
| DAX measures | ~20 governed measures (On-Time %, In-Full %, OTIF %, 3M rolling, MoM delta, anomaly flag, supplier priority score, …) driven by two What-If parameters |
| Report | Single PBIR page with KPI cards, trend line, supplier scatter, failure-breakdown bar, and an HTML narrative callout for the Oct 2025 anomaly |
| BPA rules | Tabular Editor Best Practice Analyzer rules enforcing model conventions |
| Theme | Custom LEGO-palette Power BI theme (red reserved for alerts) |

---

## How it was built — the agentic approach

The entire project was driven from the **Claude Code** terminal/desktop app with no manual edits to TMDL or PBIR JSON. The key ingredients:

### MCP servers
- **`powerbi-modeling-mcp`** — connects Claude directly to Tabular Editor's Analysis Services endpoint; used to write and validate DAX measures live against the model
- **`pbir` CLI** (`pbir.tools`) — PBIR-aware report editor; used to add/modify visuals, apply theme, manage bookmarks, and run `validate` / `backup` before bulk edits

### Claude Code Skills / Plugins
- **`data-goblin/power-bi-agentic-development`** — a plugin that adds domain-aware skills to the session:
  - `pbip` — PBIP project structure, cross-cutting renames, schema validation
  - `tmdl` — direct TMDL file authoring and syntax guidance
  - `connect-pbid` — TOM / ADOMD.NET patterns for querying PBI Desktop's local AS instance
  - `bpa-rules` — interactive BPA rule generation for Tabular Editor
  - `tabular-editor` — C# scripting, TE2 CLI automation
  - `otif-conventions` and `dax-style` — project-specific guardrails kept in `.claude/skills/`
- **`pbir-format`** skill — PBIR JSON schema reference used when authoring visual configs

### Workflow
1. Spec defined upfront in `context/OTIF_Dashboard_Spec.html` (question → visual → measure → action)
2. Claude generated the Python data script, the TMDL star schema, and all DAX measures
3. Every generated measure was executed via `connect-pbid` against PBI Desktop before committing
4. Report visuals were built via `pbir` tool calls; Claude validated the JSON after each change
5. Git commits were made in small steps throughout, with TMDL diffs reviewed as code

---

## Pre-requisites to run locally

| Requirement | Notes |
|---|---|
| Power BI Desktop | Free; open `OTIF.pbip` after updating the CSV paths (see below) |
| Python 3.10+ | Only needed to regenerate synthetic data |
| Tabular Editor 3 | Optional; needed to apply BPA rules in `OTIF.SemanticModel/bpa-rules.json` |
| Claude Code | Only needed if you want to iterate on the model/report agenically |

### Update CSV data source paths

The `.tmdl` partition sources use placeholder paths (`C:\path\to\repo\data\...`). After cloning:

1. Open each file in `OTIF.SemanticModel/definition/tables/` and replace `C:\path\to\repo` with the full path to your local clone.
2. Open `OTIF.pbip` in Power BI Desktop and refresh.

Alternatively, regenerate the CSVs with:

```bash
python data/generate.py
```

---

## Project structure

```
OTIF.pbip                          # PBIP entry point
OTIF.SemanticModel/                # Semantic model (TMDL)
  definition/
    tables/                        # One .tmdl per table + measures
    relationships.tmdl
    model.tmdl
OTIF.Report/                       # Report (PBIR)
  definition/
    pages/                         # One folder per page
  CustomVisuals/                   # HTML renderer visual
  StaticResources/                 # Custom theme JSON
data/
  generate.py                      # Synthetic data generator
  *.csv                            # Pre-generated CSVs (committed for convenience)
context/
  OTIF_Dashboard_Spec.html         # Design brief / measure library
  LEGO_SupplierServices_Theme.json # Power BI theme
```

---

## Honest assessment

The agentic loop worked well for the model layer — DAX was generated, validated against live data, and committed without any manual Desktop interaction. The TMDL authoring via MCP + skills was the most impressive part: measures came out correctly structured on the first or second attempt, and the BPA rules enforced conventions throughout.

The report layer was more iterative. The `pbir` tooling kept the JSON valid, but translating a design intent into the right PBIR property combinations took several back-and-forth rounds.

**The end result was functional but not visually polished.** The layout achieves its analytical purpose — KPIs, trend, supplier ranking, failure breakdown, anomaly callout are all there — but I didn't spend much time on design prompting, spacing, or colour tuning. With more deliberate design instructions the output would look considerably sharper. The tooling is capable of it; the bottleneck here was prompt effort, not the tools.

---

## Disclaimer

This is a portfolio / educational project. All supplier names, data, and scenarios are entirely fictional. The LEGO name is used only to frame the domain context and implies no affiliation with or endorsement by the LEGO Group.

end-to-end-pbi-report-w-AI

end-to-end-pbi-report-w-AI Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

end-to-end-pbi-report-w-AI

end-to-end-pbi-report-w-AI Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

Download README