All projects

Homestar: Las Vegas Site Selection

Analytics EngineeringData Engineering
Screenshot of Homestar: Las Vegas Site Selection

Homestar answers the question a small business asks before signing a lease: which of these sites is best? It draws trade areas as 1, 3, and 5 mile rings or as 5, 10, and 15 minute drive times from a self-hosted routing engine, then scores each site from Census, employment, OpenStreetMap, traffic count, transit, and county zoning data, all free public sources (only Census needs a free key). Every composite score breaks down into raw value, percentile against the metro, direction, weight, and contribution, and the contributions add up to the score. Missing data is excluded and the weights rebalanced rather than counted as zero, and the retail profiles' default weights follow published site-selection research. The demo is a fixed study: a comic shop choosing a second location among three Las Vegas sites.

What it proves

A multi-source public-data warehouse turned into a decision product whose scores can be defended line by line.

How it was built

Claude Code, test-first, every modeling choice signed off and merged through pull requests, with CI that builds the dbt warehouse from seeds and runs its data tests without any keys.

What is real, and where I was learning

What is real: all four signal families score on real public data, with drive-time trade areas, six business profiles, a weight-sensitivity analysis, and the demo deployed. Where it stops: rent is left out because there is no free feed, zoning covers Clark County and the City of Las Vegas only, and orthodontist competitor counts are a floor because OpenStreetMap tagging is incomplete. The repo is private.

PythondbtDuckDBStreamlitPyDeckValhallaRailway