← All case studies

Scalekrut · Forge

A skill taxonomy as a graph, not a keyword list.

01

The situation

A recruitment marketplace needed to match people to roles where the useful signal was adjacency — this person hasn't done that exact job, but everything they have done sits one hop away from it. Relational schemas make that query expensive and the results shallow.

1

Graph-native skill taxonomy

2

Sides of the marketplace

0

Keyword-match fallbacks

PythonPandasData eng.AirflowSparkAnalytics eng.dbtONE HOP AWAY IS STILL A MATCH

02

What we built

We modelled the skill taxonomy as a property graph, so adjacency became a traversal instead of a join, and built the marketplace around it: candidate and employer flows, matching, and the operational tooling to curate the taxonomy as it grew.

Not handed over.

03

Where it stands

The company ceased operations when its funding round did not close. The platform was in production and functioning at the time; the outcome was a funding outcome, not a product one. Named here with permission.

Handover status — Not handed overThere was no handover, because the company wound up. The Mantle was built and in production; the outcome was a funding one.

Neo4j · Node.js · React · PostgreSQL · Python

Next

Forge the model. Build the system. Own it.

Book a 15-min architecture call