Home
Data EngProof of Work

Healthcare Analytics on Microsoft Fabric

Power BI you can code-review — openFDA drug-safety on Microsoft Fabric, deployed through the API with sub-1.3s DAX and counts that reconcile against GCP.

View Code
💊 300
Real openFDA reports
⚡ 1.15s
DAX p50 (all < 5s)
🧱 Direct Lake
Semantic model as code
🔗 GCP=Fabric
Counts reconcile
Healthcare Analytics on Microsoft Fabric — demo

Tech Stack

Microsoft FabricOneLakePower BIDirect LakeDAXDelta LakePythonFabric REST API

Before / After

Before

Before

Dashboard-only BI
Data
↓
Manual Power BI
↓
Pretty chart
↓
No lineage
×

Measures live inside a .pbix binary nobody can review.

×

No tests run before the data reaches a chart.

×

No API layer — the numbers can't be reused downstream.

×

'Trust me it's right' instead of 'here's the file that proves it'.

After

After

Code-first Fabric stack
openFDA → OneLake
↓
Direct Lake model
↓
API-deployed report
↓
DAX < 5s + reconcile
✓

Power BI Ships From CodeThe Direct Lake semantic model and the report (openFDA Drug Safety Report, id ef468dc5, 3 visuals) deploy through the Fabric REST API — not by clicking through Power BI Desktop.

✓

Questions Return In SecondsReal DAX runs through the Power BI engine: drug leaderboard p50 1.15s, KPI card p50 0.66s — every query under 5s, measured (proofs/powerbi_dax_latency.json).

◐

Fabric Can't Quietly Change The Truth · MEASUREDfact + dim_drug/reaction + mart counts reconcile GCP ↔ Fabric exactly (300 = 300); the openFDA fact contract is the same one GCP and AWS load.

○

Rendered Screenshot Has A Cap · BOUNDARYExportToFile (report → PNG via API) is blocked on the free FTL trial, so the native screenshot needs a logged-in browser; the published chart is generated from the same DAX data.