QSR Site Intelligence QSR Site Intelligence

Overview

Built a multi-brand site-selection platform for a quick-service restaurant group that scores every city block in its markets, estimates store sales and shows where to open next without cannibalizing existing stores.

Highlights

1. 01

   Scaled the platform from one brand in two cities to three brand sections over a universe of 63,724 city blocks in 117 municipalities, serving Site Score, H3 competition layers and per-block sales estimates from versioned, SHA-256-checked model packages.

2. 02

   Shipped an inverse site search that ranks every candidate block, excludes those within a configurable radius (default 1.5 km) or a 5/10/15-minute drive-time isochrone of own stores, and merges winners into zones with 300 m non-maximum suppression.

3. 03

   Integrated the v1.2 log-linear sales model, which fixes four documented root causes of the prior version's error and cuts leave-one-out MAPE from 106% to 22%, and added per-brand account scoping, simulated openings and LaTeX PDF one-pagers.

Stack

• Next.js 16
• React 19
• TypeScript
• Leaflet
• h3-js
• Python
• GeoPandas
• OSMnx
• PostgreSQL
• Log-Linear Regression (LOOCV)
• Mapbox Isochrones
• LaTeX