The collection
Each one a problem worked into a structure people could trust and explain, with its reasoning intact.
Training a Model With Data That Didn't Exist
Generates a controlled, diverse set of invoices to train a document-processing model when no real dataset is available.
Provider and Location Routing Without Relational Support
Synthesized a join key within a platform that didn't support joins to route patient feedback to the correct review page.
Structured Synthetic CX Data for Testing and Demos
Generates structured, relational datasets that simulate real-world customer behavior for testing and demos.
How Over-Automation Broke Trust in an AP Team
Restores trust in a working AP system by fixing the workflow around it, not the system itself.
How Persistent State Broke Attribution
Prevents cross-system data contamination by enforcing interaction-level state instead of shared contact data.