professional-work

15

Two ways to read a wallet graph

In a wallet analysis prototype, I used three columns to show fund direction and amount, and an expandable force-directed graph to explore address relationships. I chose two libraries around each task and its implementation cost rather than standardizing on one.

Direction and amount

ABXCD12.0 ETH4.5 ETH9.0 ETH2.5 ETH

Incoming, central, outgoing

Address relationships

ABXCD

A force layout for unfolding connections

The same addresses arranged for two different tasks. Data and layouts are illustrative.

An explanatory diagram created with fictional data to describe professional experience. No original company screens or code are included.

One dataset, two kinds of exploration

Money Flow needed to show the direction and amount of funds entering and leaving a central address. Address Clustering needed to let a user unfold connections and explore the structure.

I used React Flow for a three-column incoming → central → outgoing layout. For address relationships, I used react-force-graph-2d for a force-directed layout with expandable connections.

I first explored using React Flow for relationships too. The expected cost of building the desired presentation led me to look for another option. Rather than standardizing on one library, I chose around the layout and interaction each view required.

Encoding direction and amount in the layout

I separated incoming and outgoing connections by flowType, then sorted each group by aggregate amount in descending order. Node counts and spacing determined their coordinates on either side of the central address. Addresses and aggregate amounts became nodes, edges, and labels.

Position and arrows conveyed direction. Labels and ordering conveyed amount.

Exploring relationships through expansion state

For Address Clustering, I converted address-string connections into node references and built neighbor, edge, and child-connection information. Starting at the central address, I traversed child connections and selected the visible graph according to expansion state.

Double-clicking expanded or collapsed connections. When an address had no child data, its mock data was added to the graph. Hovering highlighted adjacent connections; dragging pinned a node’s position. Canvas rendered the icons and address labels.

What was implemented, and what was not validated

Over one month in July 2024, I implemented the entire frontend, connecting library selection, mock-data transformation, layouts, and exploration interactions in two graph views. I used Electron and electron-builder to run the React frontend as a desktop application and produce distributable packages.

The work used mock data. Performance and usability on real production data were not validated. Search and filtering supported the visualizations rather than being the focus of the work.

The key decision was to let the information a user needed to read determine the transformation, layout, and interaction—and to consider implementation cost alongside library capabilities.