Barcode-based apps (Yuka, Fooducate) fail on foreign, regional, or unpackaged food - no barcode, no data. Zukr instead runs on-device OCR (Apple Vision, .accurate + language correction) directly on the camera feed, matching against a database of 65 sugar aliases across 18 languages, and draws bounding boxes over matches in real time.Hardest part: getting word-level bounding boxes right from Vision s line-level text observations - no undocumented APIs, no manual position estimation, just candidate.boundingBox(for:).Everything runs offline. Live on iOS; Android build is ready and heading to Google Play. Built solo in Swift/SwiftUI/AVFoundation. Would love feedback, especially on OCR accuracy on tricky labels.