On-Device Computer Vision, ML Systems & Reliable Image Uploads at Syarah
Conceived, planned, and delivered a lightweight, on-device ML car-photo classification system using a custom-trained 42-class MobileNetV3 LiteRT/TFLite model. Integrated across bulk upload, image validation, and live camera workflows, it significantly reduced photography time and human classification errors.
Built a real-time Android VIN scanner using CameraX, LiteRT, and ML Kit, with a lightweight custom-trained quantized detector and a complex VIN recognition algorithm covering OCR validation, checksum repair, and character-confusion recovery.
Also developed a reliable upload manager that substantially reduced image loss during transfers.