Fabric and OneLake lakehouse
Fabric positions OneLake as a unified lake foundation and supports lakehouse patterns for enterprise analytics.
Microsoft LearnAzure student outcome intelligence platform
Governed dropout-risk architecture for a Norwegian university.
A professional case-study package built around Microsoft Azure, Fabric, Purview, Azure ML, and Power BI. The solution unifies SIS, LMS, ERP, and campus signals into point-in-time risk bands advisors can explain and audit.
AzureFabric, ADLS Gen2, Azure ML
PurviewCatalog, lineage, governance
RLSAdvisor-scoped access
RAIFairness and explainability review
Case Summary
The brief is a 20-30 minute platform design for early dropout-risk detection at a large Norwegian university.
One-line answerBuild a governed Azure data spine and advisor queue, not just a dropout model.
What must be defensibleIdentity resolution, time-aware student history, leakage-safe features, calibrated capacity bands, fairness checks, reason codes, and access audit.
Institution35,000 students across 8 faculties, with diverse study modes, programmes, and engagement patterns.
Fragmented dataSIS, LMS, ERP, and campus systems use different identifiers, formats, update frequencies, files, APIs, and events.
GoalFlag non-continuation risk before grades, withdrawal, or official status records make the problem visible.
ConstraintsThe solution must be accurate, explainable, fair, auditable, privacy-aware, and stable after launch.
Azure Blueprint
Concrete service choices for the brief: mixed ingestion, governed lakehouse, reproducible ML, and restricted advisor delivery.
IngestAzure Data Factory or Fabric Data Factory for APIs/files, with Event Hubs for high-frequency LMS and campus streams.
StoreADLS Gen2 raw evidence and Microsoft Fabric OneLake lakehouses for bronze, silver, and gold data products.
GovernMicrosoft Purview lineage, Entra groups, Key Vault secrets, Azure Monitor alerts, and immutable access audit tables.
ActAzure ML model registry and Responsible AI review feeding a Power BI/Fabric advisor queue with row-level security.
Microsoft Reference Basis
The Azure choices are grounded in current Microsoft architecture and product documentation.
Medallion data layers
Bronze, silver, and gold layers progressively improve data quality from raw evidence to trusted analytical products.
Microsoft LearnEvent Hubs Capture
High-frequency LMS and campus events can be captured into Blob Storage or ADLS for long-term retention and batch processing.
Microsoft LearnPurview lineage
Data Factory lineage can be connected into Microsoft Purview so source, transformation, and output metadata are auditable.
Microsoft LearnAzure ML Responsible AI
Responsible AI dashboards can attach interpretability, error analysis, and review artifacts to registered models.
Microsoft LearnPower BI row-level security
Power BI RLS restricts model rows for Viewer users and supports Microsoft Entra security groups for advisor access.
Microsoft LearnFormula Spine
The core math behind the solution, formatted for review and interview discussion.
Risk
ri,t = P(Yi,t+h = 1 | Xi,t)
Probability that student i will not continue in the next eligible term.
Leakage
usable(record, t) = 1[event_time ≤ t and available_at ≤ t]
Only records known at the scoring date can enter the feature snapshot.
Capacity
τred = quantile(r, 1 - Cred / N)
Risk thresholds are tied to how many students advisors can realistically support.
Fairness
gap = max(metricg) - min(metricg)
Governance tracks recall, FPR, calibration, and flag-rate gaps across groups.
Main Deliverables
Primary materials to review or present directly from the browser.
Written Solution
Azure-based answer with assumptions, service blueprint, architecture, data model, ML approach, fairness, audit, roadmap, and trade-offs.
Presentation Deck
Self-contained slide deck for a 20-30 minute walkthrough, plus the exported PDF.
Demo Dashboard
Generated dashboard showing dropout-risk distribution, advisor lists, fairness checks, validation metrics, and audit samples.
Case Prompt
Original case PDF and extracted prompt text used to shape the solution package.
Architecture Reference
Standalone Azure/Fabric SVG architecture diagram with ingestion, lakehouse, ML, advisor delivery, and control-plane layers.
Generated Outputs
CSV outputs for advisor workflow, features, validation, fairness, data dictionary, and access audit review.
Evidence Views
Charts generated by the prototype and included in the dashboard.
Generated Data And Reports
Artifacts produced by the demo pipeline and available from the site.
Advisor risk list
Student-level advisor queue with risk scores and reason codes.
Open CSV
Feature and predictions table
Point-in-time weekly feature set with model outputs.
Open CSV
Fairness report
Flag-rate checks and cohort comparisons for governance review.
Open CSV
Model validation metrics
Summary validation metrics from the prototype run.
Open CSV
Data dictionary
Field definitions for source samples, features, predictions, and outputs.
Open CSV
Access audit sample
Example governance log for role-based access and auditability.
Open CSV
Source Samples And Scripts
Prototype source data samples and local rebuild scripts.
Source Sample Tables
SIS, LMS, finance, enrollment, status, campus activity, and identity-map sample data.
Rebuild And Server Scripts
Python generators for the case view and demo outputs, plus the Node server used by this site.