Halmahera

(https://streetproof.halmahera.site/en) StreetProof turns facility damage reports into assigned repair work that only closes with photo proof. In most buildings, a tenant reports a leak in a chat group, someone says "on it", and nobody can later show whether it was fixed. StreetProof gives every building a QR code. A tenant scans it, picks the location, describes the damage by typing or by voice, adds a photo, and sends it without creating an account. They keep a private link that shows progress in plain stages. The building team works through roles. Operators triage a queue, technicians accept work and submit completion proof (a category checklist plus after photos), and supervisors verify or send the proof back with a written reason. Admins manage buildings, locations, members, QR posters and per-site time zones (WIB, WITA, WIT), and see where work waits on a dashboard. Every status change is recorded, and a verified repair can be reopened. Meta's models help at each hand-off but never decide: - Muse Voice Transcribe turns a tenant's voice note into text they check before sending. - Muse Spark drafts the ticket: summary, category, priority, the reason, what context is missing, and possible duplicate reports, each tied back to the report and photos. - SAM 3.1 outlines the reported damage on the photo as a pointer, not a measurement. - Muse Spark compares before and after photos and tells the supervisor whether they show the same place and whether the problem is still visible. It cannot accept or reject proof. Every suggestion is labelled "Not verified". The operator confirms the triage, and only a supervisor with a photo can close work. StreetProof also runs as a Telegram bot for reporters, technicians and supervisors, sends deadline reminders and escalations, exports the queue as CSV, and has rate limits, data retention and off-site backups. It is deployed at https://streetproof.halmahera.site with synthetic demo data (password for all demo accounts: streetproof-demo). Stack: Next.js, a Node worker and a Telegram bot in a TypeScript monorepo, PostgreSQL, S3-compatible storage (Cloudflare R2), Docker Compose on Coolify.