Live: https://findyourvote.org · Open: 01_live-dashboards/findyourvote-determinism-report.html
Problem. Voters can't easily see which candidates actually match their priorities, backed by real records rather than spin.
What I built. A neutral, shipped platform: enter an address → district lookup → set issue positions (-5..+5) and weights → get ranked candidate matches scored from 25+ public data sources (Census Geocoder, Congress.gov, Open States, OpenFEC, GDELT, Wikipedia, Google Fact Check, GovTrack). Express + better-sqlite3 (dev) / Supabase (prod), Vercel + GitHub Actions, Helmet/CSP/rate-limit hardening, 46 passing tests. Real data: 572 candidates, 664 races, 362 officials.
Featured artifact — the determinism report. A self-contained HTML audit documenting that the scoring pipeline is repeatable and atomic: schema, ingestor logic, caching strategy, and honest ceilings. It's the "show your work" companion to the live site.
Why it matters. Large-scale multi-source API orchestration plus an explicit determinism/repeatability audit — the reliability mindset an applied engineer brings to data-heavy systems.
Find Your Vote is live at findyourvote.org. You put in your address, set how much you care about different issues, and it ranks candidates by how well their actual record matches, scored from twenty-five-plus public data sources. What I'd point to for an engineering interview is this determinism report: I audited my own pipeline to prove the scoring is repeatable and atomic, documented the schema, the caching, and the honest ceilings of the data. It's me holding my own system to a reliability bar, which is what you'd want me doing to yours.
GG · Applied AI Engineering portfolio · synced to youdontneedmy.help design