Skip to content

Latest commit

 

History

History
103 lines (86 loc) · 7.56 KB

File metadata and controls

103 lines (86 loc) · 7.56 KB

Microsoft Fabric for Healthcare Providers — One-Day Hands-On Lab

Unified Patient Intelligence Platform

Build an end-to-end analytics solution on Microsoft Fabric that helps a hospital system improve patient outcomes, reduce costs, and meet regulatory requirements.


🏥 Who Is This Lab For?

Healthcare providers — hospital administrators, clinical informaticists, data analysts, IT leaders, and quality improvement teams — who want to understand how Microsoft Fabric can unify their data platform.

⏱️ Agenda

Time Module Duration Fabric Capability
9:00 – 9:20 Module 0 — Introduction & Healthcare Context 20 min Overview
9:20 – 9:50 Module 1 — Lakehouse & Data Ingestion 30 min Lakehouse, Upload, Spark Notebook
9:50 – 10:35 Module 2 — Data Engineering 45 min Spark Notebooks, Delta Tables
10:35 – 10:45 Break 10 min
10:45 – 11:30 Module 3 — Semantic Model, Dashboard & Power BI Copilot 45 min Semantic Model, Power BI, Copilot
11:30 – 12:30 Module 4 — Real-Time Analytics 60 min Eventhouse, KQL, Real-Time Dashboard
12:30 – 1:15 Lunch Break 45 min
1:15 – 1:55 Module 5 — Data Agent 40 min Prep Data for AI, Fabric Data Agent
1:55 – 2:25 Module 6 — Gen AI: Clinical Intelligence 30 min Notebooks, Fabric AI Services
2:25 – 3:25 Module 7 — Predicting Readmissions with AutoML 45–60 min Azure OpenAI, FLAML AutoML, MLflow
3:25 – 3:40 Wrap-Up & Q&A 15 min
(Optional) Module 8 — Building with VS Code Agent Mode 60–90 min GitHub Copilot Agent Mode, VS Code
(Optional) Module 9 — Data Agent Evaluation (LLM-as-Judge) 30 min Data Agent SDK, LLM-as-Judge, Semantic Link
(Optional) Module 10 — Fabric IQ: Ontology & Plan 45–60 min Fabric IQ, Ontology, Plan, Data Agent
(Optional) Module 11 — AI Search + Data Agent (Structured + Unstructured) 45–60 min Azure AI Search, Fabric Data Agent

Total core day: 5.5 hours (9:00 AM – 3:30 PM including breaks and lunch). Optional modules can be done independently after the lab.

📂 Repository Structure

FabricHackathon/
├── README.md                          ← You are here
├── lab-guide/
│   ├── Module00_Introduction.md       ← Healthcare challenges & lab overview
│   ├── Module01_Setup_and_Data_Ingestion.md  ← Lakehouse + Data Pipeline alt path
│   ├── Module02_Data_Engineering.md
│   ├── Module03_Semantic_Model_and_Dashboard.md  ← + Power BI Copilot
│   ├── Module04_RealTime_Analytics.md
│   ├── Module05_Data_Agent.md         ← Prep Data for AI + Data Agent
│   ├── Module06_GenAI_Clinical_Intelligence.md ← AI-powered clinical note analysis
│   ├── Module07_Readmission_Prediction.md ← GenAI-powered readmission prediction with AutoML
│   ├── Module08_VSCode_Agent_Mode.md  ← (Optional) Build entire lab with AI agent
│   ├── Module09_Data_Agent_Evaluation.md ← (Optional) LLM-as-Judge evaluation pipeline
│   ├── Module10_Fabric_IQ.md          ← (Optional) Ontology + Plan with LLM-assisted design
│   └── Module11_AI_Search_and_Data_Agent_Integration.md ← (Optional) Combine unstructured + structured responses
├── data/
│   ├── generate_healthcare_data.py    ← Python script to regenerate data
│   ├── ai_search_sample_docs/          ← Sample unstructured docs for Azure AI Search (Module 11)
│   ├── patients.csv                   ← 200 synthetic patients
│   ├── encounters.csv                 ← ~1,000 hospital encounters
│   ├── conditions.csv                 ← 428 diagnoses (ICD-10 coded)
│   ├── medications.csv                ← 640 medication records
│   ├── vitals.csv                     ← 3,800+ vital sign readings
│   ├── clinical_notes.csv             ← 150 clinical notes (for Gen AI)
│   └── claims.csv                     ← ~1,000 billing/claims records
├── notebooks/
│   ├── 01_Bronze_Data_Ingestion.py    ← With FHIR mapping & data exploration
│   ├── 02_Silver_Transformations.py   ← With ICD-10 & SIRS clinical context
│   ├── 03_Gold_Analytics.py           ← With KPI business rationale
│   ├── 04_GenAI_Clinical_Notes.py     ← With prompt engineering explanations
│   ├── 05_RealTime_Vitals_Simulator.py  ← With patient archetype & SIRS docs
│   ├── 06_Predictive_Readmission_Model.py ← GenAI feature engineering + XGBoost
│   └── 07_Data_Agent_Evaluation.py    ← LLM-as-Judge evaluation pipeline
└── resources/
    ├── kql_queries.kql                ← KQL queries for real-time analytics
    └── dax_measures.md                ← DAX measures for semantic model

🚀 Prerequisites

  • A Microsoft Fabric workspace (capacity F64 or higher recommended)
  • A web browser (Microsoft Edge or Google Chrome)
  • The synthetic CSV data files from the data/ folder (pre-generated)
  • For Module 6 (Gen AI): An Azure OpenAI Service endpoint with a GPT-4o deployment

📊 What You Will Build

By the end of this lab, you will have:

  1. A Lakehouse with Bronze → Silver → Gold data layers containing clinical, operational, and financial healthcare data
  2. Spark Notebooks that compute hospital quality measures including 30-day readmission rates, average length of stay, and ED utilization
  3. A Semantic Model (star schema) with measures for readmission rate, bed occupancy, and revenue analysis
  4. A Power BI Dashboard with pages for Patient Volume & Flow, Quality & Readmissions, and Population Health — plus Power BI Copilot testing
  5. A Real-Time Dashboard monitoring simulated patient vitals with sepsis early-warning detection
  6. A Data Agent that lets you ask questions about your healthcare data in natural language, with Prep Data for AI configuration
  7. A Gen AI Notebook that summarizes clinical notes and suggests ICD-10 codes
  8. A Readmission Prediction Model using Gen AI-assisted feature engineering and AutoML (FLAML)
  9. (Optional) VS Code Agent Mode — experience building the entire lab through conversational AI with GitHub Copilot
  10. (Optional) Readmission Prediction Model — Gen AI-assisted feature engineering with XGBoost to predict 30-day hospital readmissions and generate patient risk scores
  11. (Optional) AI Search + Data Agent Integration — combine Azure AI Search document context with Fabric metrics in one grounded response

⚠️ Important Notes

  • All data in this lab is 100% synthetic. No real patient information (PHI) is used.
  • The data was generated using a Python script (data/generate_healthcare_data.py) with realistic distributions but fictional names and records.
  • This lab is for educational purposes and demonstrates Fabric's capabilities. Production healthcare solutions require additional security, compliance (HIPAA), and governance controls.