Skip to content

README.md

Latest commit

 

History

History
57 lines (47 loc) · 3.44 KB

README.md

File metadata and controls

57 lines (47 loc) · 3.44 KB
Raw

AML Dashboard — Power BI Project

A .pbip project (Power BI's text-based project format), not a .pbix file. Open AML_Dashboard.pbip in Power BI Desktop (2024+ recommended — the TMDL/PBIR formats need a reasonably current Desktop build).

What's real vs. what needs your hands

Fully built and should load as-is: the semantic model (AML_Dashboard.SemanticModel/) — 5 real tables wired to aml_dev's actual Gold/Silver schema via the Databricks connector (using this workspace's real SQL warehouse endpoint, dbc-08add949-9c19.cloud.databricks.com / warehouse 59901b31d31db40a), 2 relationships, and a _Measures table with 13 DAX measures covering every KPI in docs/00_business_charter.md.

Built, but unverified in Desktop — read this before assuming anything below "just works": the report (AML_Dashboard.Report/) has 3 pages with 19 real visuals hand-authored directly as PBIR JSON (cards, tables, a column chart, a donut, a line chart, slicers) — not just an empty shell. Every file is valid JSON (checked directly), and every field/measure reference matches something that actually exists in the semantic model. What's not verified is whether Power BI Desktop's current schema for visualContainer JSON matches what was hand-written here field-for-field — that can only be confirmed by actually opening it, which wasn't possible in the environment this was built in.

If it opens clean: you have a working 3-page dashboard already laid out — check it against DASHBOARD_BUILD_SPEC.md for anything you'd still want to add or restyle.

If Desktop offers to "repair" the report, or specific visuals don't render: the .SemanticModel folder is completely independent of the report and should be unaffected either way. Two fallback levels, cheapest first:

  1. Delete the one malformed visual's folder (pages/<page>/visuals/<visualId>/) and rebuild just that visual by hand — PBIR's one-file-per-visual layout means a bad visual shouldn't take the rest of the page down with it.
  2. If the whole report won't open: in Desktop, File → New report connected to AML_Dashboard.SemanticModel, then rebuild pages from DASHBOARD_BUILD_SPEC.md (a few minutes of drag-and-drop per page — the spec was written to stand on its own regardless of whether the hand-authored JSON survives contact with real Desktop).

First-time setup in Desktop

  1. Open AML_Dashboard.pbip.
  2. On first refresh, Desktop will prompt for the Databricks connection — sign in with whatever auth method you use for meridian-dev (Azure AD / personal access token).
  3. WatchlistMatchSummary's jurisdictions column (an ARRAY<STRING> in Delta) was deliberately left out of the model query — Power Query surfaces array columns as list values that need a one-time Extract Values or Expand to New Rows step in the Power Query Editor. The M code in WatchlistMatchSummary.tmdl has the exact spot to add it back if you want it.

The ROI number is a documented assumption, not a measurement

Estimated Analyst Hours Saved (in _Measures.tmdl) uses a hardcoded MinutesPerLowPriorityReview = 8 constant. There's no real investigator feedback loop in this portfolio project to calibrate that number from — it's a placeholder assumption, labeled as such in the DAX comment. Say so out loud if this comes up in an interview; presenting an assumption as a measured result is the opposite of the explainability this project is built around.