Actuarial studies · 2025
Health Index
A clearer view of where healthcare services are needed.
Understanding where healthcare services are concentrated starts with a consistent view of the underlying data. This project brings together business locations, services and estimated staffing, then links them to population data. It gives local supply-and-demand analysis a clearer starting point.
- Status
- Production
- Domain
- Health economics · Geospatial ML
- Role
- Pipeline architecture, modelling
- Year
- 2025
Explore the workflow with illustrative content.
Start with the place
The base layer establishes the streets and area around a sample location. Select a location below to move the focus.
Add the surrounding context
A separate catchment layer shows how the surrounding area relates to the selected place. Spread the layers to see how they fit together.
Compare the nearby activity
Sample sites and a route sit above the catchment. Switch between locations to see the area of interest move across the map.
Illustrative content · no live data
Connect records with the places they describe
Stack
- PostgreSQL
- Python
- Playwright
- LLM extraction
- scikit-learn
- Geospatial · SA2
A healthcare data pipeline that combines service directories, website research and staffing estimates to support local supply-and-demand analysis.
Bring scattered information together
A public services directory provides the starting register.
Additional discovery and website research fill in gaps, while matching rules reconcile duplicate names and locations. The pipeline keeps these stages separate so an incomplete run can be resumed and a questionable record can be traced back.
Make the data useful locally
Each site is linked to statistical areas, allowing the register to join with census and demographic data.
That makes local comparisons possible: where services are concentrated, how the surrounding population differs and which areas warrant a closer look.
Estimate what is rarely published
Language-model extraction turns website text into structured facts about services and practitioners.
A separate supervised model estimates full-time-equivalent staffing where it is unavailable. Keeping extraction and estimation separate makes their different sources of uncertainty easier to review.
A foundation for the next tools
The index supplies a common data foundation for workforce modelling, dashboards and site analysis.
Its usefulness depends on maintaining the register as businesses change, so repeatable updates, entity matching and geographic joins are central parts of the design.
Pipeline
From directory to local supply analysis
Stage Can reject
-
01
Directory
Baseline register
-
02
Geocode
Statistical areas
Ready for census joins
-
03
Discover
Additional candidate sites
-
04
Resolve
Matched clinic records
-
05
Extract
Structured facts
LLM reads prose only
-
06
Predict FTE
Staffing estimate
Supervised estimation
Behind the work
The skills
that came together.
Data engineering
- Directory ingestion and website enrichment
- Entity matching and deduplication
- Repeatable PostgreSQL pipelines
Analysis
- Structured extraction from website text
- Supervised staffing estimation
- Census joins and local supply analysis
Project highlights
-
A shared register
Directories and research brought into one structure
-
Ready for local analysis
Clinic records linked to census geography
-
Staffing estimates
A supervised model for an often-missing input