Team Nihan

Every renovation quote in Pakistan and Saudi Arabia starts the same way: a contractor drives to the site, measures every room by hand, and comes back days later with a number nobody can check. It takes about an hour per room, the quotes vary wildly, and contractors lose unpaid hours measuring jobs they never win. Mimaar AI replaces that site visit with a few phone photos. Photograph a room, pick the work you want priced, and within seconds Mimaar outlines every wall, floor, ceiling, door, window, crack, damp patch and peeling surface. It measures them in square metres and gives a cost range in SAR, PKR or USD. A written report follows in English, Arabic or Urdu, ready to export as a branded PDF quote. No tape measure needed: it estimates from a standard door and typical room sizes, and clearly labels what is estimated. HOW IT USES META’S MODELS Mimaar chains two Meta open-weight models, running entirely on one consumer laptop GPU (8 GB). No cloud, no API key, and photos never leave the device. SAM 3 segments ten building elements and defects per photo in about a second. Muse Glimmer 30B reads the photos and the measurements, decides which defects are real and how serious they are, and writes the report in three languages. The same code switches to Muse Spark and SAM 3.1 on the Meta Model API with one setting. We built the local path because the API is not available in our region, and it turned into our biggest strength: private and free per estimate. WHAT MAKES IT DIFFERENT The AI never invents a number. SAM 3 sees and Muse Glimmer judges and writes, but every count, area and cost is calculated by code and validated. Each finding shows a “Why this?” trace, so a quantity surveyor can check the work instead of redoing it. A 30B model made practical on an 8 GB GPU. An instant report comes from SAM 3 and code, and Glimmer’s write-up streams in afterwards. Speculative decoding and a tightly scoped prompt cut a write-up from 11 minutes to under one. Built for real rooms. It combines up to six photos of one room, ignores furniture when measuring walls, doesn’t mistake wardrobe doors for room doors, merges duplicate defects, and refuses photos that aren’t rooms. Users can tap to correct any detection and the cost updates instantly. Ready for business. Contractors import their own price file with their rates, logo and contact details, and every report becomes their branded quotation. PROVEN ON REAL PHOTOS On real Pakistani property photos, Mimaar identified 20 of 22 elements correctly, including every door, window, floor and ceiling, with zero false defects. The instant report took 2 to 4 seconds per photo. The project includes 80+ automated tests. WHO IT’S FOR Finishing contractors and site engineers who need same-day quotes, quantity surveyors checking estimates, and property agents describing a home’s condition, in the languages their clients actually speak. From an hour with a tape measure to a few photos and a few seconds. Built entirely during the Meta Global AI Developer Hackathon (Oct 27 to Nov 4). Mimaar AI is decision support, not a structural or safety assessment.