Challenge
Migration advice is scattered across government pages, forum threads, and agents who are paid whether or not you get in. It goes stale quietly: a fee changes, a quota fills, a route closes, and the version circulating is still last year's. The cost of believing it is not an inconvenience — it is a year of someone's life. Most people who lose that year were eligible for something. Nobody told them the truth early enough, in a language they read, with the numbers as they actually stand today.
Approach
Eligibility is computed in tested, auditable code rather than delegated to a model — the rules are the product, so they have to be inspectable and they have to be right. AI sits around that core, not inside it: two models independently verify each rule against its official government source, and a human signs off before anything publishes. Every statistic on the site carries a link to the page it came from and the date it was last checked, so a reader can audit the claim instead of trusting it. Routes are scored 0–100 across eligibility, affordability, timing, and refusal risk; costs are itemised and converted to the user's currency with estimates explicitly labelled as estimates; timelines are calculated backwards from real intake dates rather than quoted as averages. Built on Next.js on Vercel with Firebase behind it, localised across English, French, and German, with uploaded documents encrypted and deletable by the person who uploaded them.
Stack