Acceptance Radar Acceptance Radar

Overview

Built a territorial payment-acceptance platform for a global payments network that scores Mexico's ~5.5M merchants with an explainable conversion-propensity model and lets teams filter, map and export prioritized merchants with coordinates.

Highlights

1. 01

   Engineered a Node ETL that extracts 5.49M POIs with payment-method flags from a read-only 45 GB Postgres warehouse into DuckDB and publishes a 70 MB Parquet+zstd serving artifact.

2. 02

   Designed a geometric propensity core (transaction value × acceptance gap × viability × neighborhood pressure) with three-state payment evidence and completeness shrinkage below 0.35 confidence.

3. 03

   Cut a percentile step that hung for over 40 minutes to 12 seconds with exact mid-rank percentiles over distinct values, and made map, list and export counts always match.

Stack

• Next.js 16
• React 19
• TypeScript
• DuckDB
• PostgreSQL
• Parquet (zstd)
• MapLibre GL
• H3
• Log-Odds Propensity Modeling
• Vercel Blob
• Zustand
• Radix UI