Mahmoud Amr
A production-grade geospatial data product: real Census geometry, a drift-resilient ETL, and 33k+ regions that recolor instantly. Built end to end by Mahmoud Amr.
What this demonstrates
33,000+ ZIP polygons recolor the instant you switch metrics — geometry is fetched once as vector tiles and never re-rendered.
Every source is validated against an explicit contract; a drifted feed is rejected with a precise diagnostic and isolated, so bad data never corrupts good data.
Data modeling, the pipeline, a typed API over PostGIS, the interactive UI, CI, and the production deploy — designed and shipped as one coherent system.
What it is
An interactive choropleth of US housing-market data across four geographic resolutions — state, metro, county, and ZIP. Boundaries are real US Census geometry; metric values are synthetic, deterministic, and clearly labeled. The point is a production-shaped geospatial data product, end to end.
Architecture
A typed pnpm monorepo: a Next.js (App Router) front end with MapLibre GL JS rendering self-hosted PMTiles, typed route handlers over Postgres + PostGIS, a shared Zod contract, and a Python ETL. Everything runs locally with one Docker Compose command and deploys to Vercel + Neon at ~$0.
The performance trick
Switching metrics never refetches or re-renders geometry — values are pushed via MapLibre feature-state and the fill expression interpolates over them. That is what keeps tens of thousands of regions instant.
Proper data modeling
A dimensional model over PostGIS — region and metric dimensions, a monthly fact table, and a materialized view for fast map shading — with a single cross-language metric registry as the source of truth.
Capabilities
Let's build something
Mahmoud Amr is a Senior Staff Engineer available for senior and staff-level engineering work — geospatial, data platforms, and AI-native products.
Data is synthetic and illustrative; geometry is real US Census TIGER/Line. Built and maintained by Mahmoud Amr.