unclaimedSF.org
Find San Francisco benefits you may be missing, with the exact rule behind every match.
Core trick
AI compiles public benefit eligibility rules offline, and the shipped app runs them as deterministic code entirely in the browser.
Description
Find San Francisco benefits you may be missing, with the exact rule behind every match. Built for SF residents with <3
README excerpt
# Public Works **Find the San Francisco benefits you're entitled to, with the exact rule behind every match.** **Live:** https://unclaimedsf.org · open-source, deployed on Vercel. Public Works screens a resident against real public-benefit rules and tells them, in plain language, which programs they qualify for, why (the exact rule that matched), and where to apply. Answer a handful of questions and learn that one enrollment (say CalFresh) cascades into several more discounts you never knew you qualified for. The defensible idea: AI does the hard reasoning offline, once, and verifies itself; the shipped app is plain deterministic code; the resident's answers never leave the browser. ## How it works: two separated halves **Half A, offline rule compilation (AI-heavy, run once).** A dynamic workflow fans out a team of subagents: one extracts each program's eligibility from its official source, then a second, independent agent adversarially verifies that extraction against the same source (catching wrong thresholds, missed clauses, invented rules). Only agreed rules are compiled into `data/ruleset.json`, which is then graded against a labeled test set. **Half B, the runtime app (deterministic, no AI).** A React app loads `ruleset.json`, generates the questionnaire from it, and runs a 3-valued match engine (`app/src/match.js`) entirely in the browser: every rule node is TRUE, FALSE, or UNKNOWN, which maps to full match, no match, or partial match. Partial matches report e