← Selected work

DigitalFUTURES · Independent interface development · 2025

Healthy Ties

Community graphs & interface design

A completed DigitalFUTURES graph-modelling project, followed by independent interface and app-layout development.

Healthy Ties cover composition
Original portrait coverOriginal portrait portfolio cover
01

Representing a community

Healthy Ties explores the facilities and walking connections around a community. The graph representation joins spatial composition and accessibility in one analytical structure.

  1. Community
  2. Walking catchment
  3. Facilities
  4. Weighted connections
Community location.
Community location.
Walking catchment.
Walking catchment.
Surrounding facilities.
Surrounding facilities.
Accessibility-based connections.
Accessibility-based connections.
02

Nodes, distances and features

Communities and facilities form nodes. Walking-network distances define connections within 1,200 metres, with Gaussian distance-decay weights. Encoded features are compressed from 768 to 16 dimensions for the model.

Edge construction, network distance and distance-decay weighting.
Edge construction, network distance and distance-decay weighting.
Encoded feature representation.
Encoded feature representation.
Compressed feature representation.
Compressed feature representation.
Modelling workflow: from feature representation to graph data.
Modelling workflow: from feature representation to graph data.
03

Comparing community subgraphs

Different facility compositions and connections produce different subgraphs. The comparison preserves multiple examples rather than reducing the study to a single network image.

Community subgraph — example 01.
Community subgraph — example 01.
01 / 12

Twelve community subgraphs shown in sequence. Pause to inspect connections and facility distributions.

04

Labels and sampling

Assessment labels are constructed from selected spatial indicators. High-, middle- and low-score ranges guide sampling; the displayed modelling approach uses binary classification. These scores characterize the assessment framework, not individual health.

Sampling across score ranges: 30% high, 40% middle and 30% low.
Sampling across score ranges: 30% high, 40% middle and 30% low.
Subgraph classification: input features, edges and assessment labels.
Subgraph classification: input features, edges and assessment labels.
Facility composition example.
Facility composition example.
A contrasting facility composition.
A contrasting facility composition.
05

Training and evaluation

Training and test curves show both learning and a remaining performance gap. Embedding plots provide another view of the representation, but do not by themselves establish generalization.

Training and test accuracy.
Training and test accuracy.
Training and test loss.
Training and test loss.
t-SNE representation.
t-SNE representation.
PCA views of labels and clusters.
PCA views of labels and clusters.
06

Spatial interpretation

The mapped output locates variation in the model-derived scores. It provides a spatial view of the assessment rather than a direct measurement of residents’ wellbeing.

Community score map from the project.
Community score map from the project.
07

Independent interface development

After the workshop, I developed the interface, app layouts and visual details. These screens translate the analytical workflow into a navigable design, connecting community selection, overview and detailed indicators.

Community selection.
Community selection.
Interface home.
Interface home.
Community overview.
Community overview.
Detailed indicator view.
Detailed indicator view.
Location and community navigation.
Location and community navigation.
Spatial-results view.
Spatial-results view.

The workshop project was completed successfully. The subsequent independent work shown here is interface design; the screens do not imply a deployed app.