Fomy IO

HoTH (Heroes of the Hay; hay, حيّ, is Arabic for neighbourhood) is a social app for reporting unsafe public spaces. A resident takes a live photo of a hazard (a broken playground ride, an open manhole, overflowing bins) and talks to an AI assistant in English or Arabic. A couple of minutes later they have a structured, prioritised ticket that explains why it is ranked where it is and is routed to the right team. Neighbours confirm reports ("I saw it too"), comment (AI-moderated), get alerts about hazards near them, and celebrate the fix once a crew's before/after photos are verified. The problem: hazards like these get walked past every day because reporting is slow, form-heavy and invisible to the people who live next to them. The people most exposed (children on a broken ride, older residents, people who would never fill in a form, or not in English) are the least likely to report. HoTH turns one photo and a short spoken conversation into a ticket a city team can act on, and makes the neighbourhood part of getting it fixed. Who it helps: - Residents: report by speaking, in Arabic or English, and see the fix happen. - Children and families: child-safety hazards are pushed up the queue by rules the model cannot override. - City teams: one clear ticket per issue, with category, outline of the damage, reasons for the priority and neighbour confirmations, instead of vague duplicate complaints. Responsible by design: - Privacy first: faces and licence plates the phone detects are blurred on the device before any photo is sent to a model. If the blur fails or takes over 10 s, nothing is sent (fail closed). A test checks that every image in every Meta request is the blurred copy, never the original. - The AI proposes, rules decide: the model suggests a priority, but deterministic safety floors can only raise it, never lower it (an open manhole is always CRITICAL; broken child equipment is at least HIGH, CRITICAL with children around). Every priority is explained with "why" chips. - Hard to abuse: live camera only (no photo library), capture time and GPS checks, screen/stock-photo detection, duplicate detection, and comment moderation. - No backend: reports stay on the phone, the API key lives in the iOS Keychain, and location sent to the model is rounded. - Honest about limits: we publish our eval including the weak spots (playground photos 87%) and a limitations section in the README. Native iOS app (SwiftUI + SwiftData). Built solo by Mosaab Yassir LAFRIMI in Abu Dhabi, UAE.